the very basic core of a man's living spirit is his

The very basic core of a man's living spirit is his passion for adventure. The joy of life comes from our

encounters with new experiences, and hence there is no greater joy than to have an endlessly

changing horizon, for each day to have a new and different sun.

― Christopher McCandless

Promoters Prof. dr. Colin Janssen Environmental Toxicology Unit (GhEnToxLab) Department of Applied Ecology and Environmental Ecology Ghent University Prof. dr. ir. Frederik De Laender Laboratory of Environmental Ecosystem Ecology Research Unit in Environmental and Evolutionary Biology Department of Biology University of Namur Dean Prof dr. ir. Marc van Meirvenne

Rector Prof. dr. Anne De Paepe

Karel VIAENE

Improving ecological realism in the risk assessment of chemicals:

Development of an integrated model

Thesis submitted in fulfilment of the requirements for the degree of Doctor (PhD) in Applied Biological Sciences

Proefschrift voorgedragen tot het bekomen van de graad Doctor in de Bio-ingenieurwetenschappen

Dutch translation of the title:

Verhoging van de ecologische realiteit in de risicoschatting van chemische stoffen: Ontwikkeling van een geïntegreerd model

Refer to this work as:

Viaene K., 2016, Improving ecological realism in the risk assessment of chemicals: development of an integrated model. PhD Thesis, Ghent University, Belgium.

ISBN 978-90-5989-891-2

The author and promoter give the authorisation to consult and to copy parts of this work for personal use only. Any other use is limited by the Law of Copyright. Permission to reproduce any material contained in this work should be obtained from the author.

TABLE OF CONTENTS

CHAPTER 1 GENERAL INTRODUCTION AND OUTLINE p. 1

CHAPTER 2 SPECIES INTERACTIONS AND CHEMICAL STRESS: COMBINED EFFECTS OF INTRASPECIFIC AND INTERSPECIFIC INTERACTIONS AND PYRENE ON DAPHNIA MAGNA POPULATION DYNAMICS

p. 19

CHAPTER 3 DEVELOPMENT OF THE DEBKISS IBM p. 39

CHAPTER 4 APPLICATION OF THE DEBKISS IBM FRAMEWORK TO ASSESS POPULATION-LEVEL EFFECTS OF COMPETITION AND CHEMICAL STRESS

p. 51

CHAPTER 5 DEVELOPMENT OF THE INTEGRATED CHIMERA MODEL p. 67

CHAPTER 6 THE CHIMERA MODEL AS A SCENARIO ANALYSIS TOOL p. 81

CHAPTER 7 GENERAL CONCLUSIONS p. 121

APPENDIX A p. 131

APPENDIX B p. 145

APPENDIX C p. 149

APPENDIX D p. 155

REFERENCES p. 157

SUMMARY p. 169

SAMENVATTING p. 173

DANKWOORD p. 177

CURRICULUM VITAE p. 181

LIST OF ABBREVIATIONS

μg microgram

CR Concentration-response

d Day

DEB Dynamic Energy Budget

DEBkiss Dynamic Energy Budget, Keep It Simple, Stupid

dw Dry weight

EC European Commission

ERA Ecological Risk Assessment

EU European Union

FOCUS FOrum for the Co-ordination of pesticide fate models and their Use

h Hour

HC5 Harmful Concentration for 5% of the species

IBM Individual-based Model

ind individuals

K Phytoplankton carrying capacity

L Litre

mg Milligram

mL Millilitre

mm Millimetre

PEC Predicted Environmental Concentration

PNEC Predicted No Effect Concentration

REACH Registration, Evaluation, Authorisation and Restriction of Chemicals

RQ Risk Quotient

SCHER Scientific Committee on Health and Environmental Risks

SSD Species Sensitivity Distribution

TKTD Toxicokinetic-toxicodynamic

WFD Water Framework Directive

ww Wet weight

1

1 GENERAL INTRODUCTION AND OUTLINE

Chapter 1

2

1.2. Ecological risk assessment

Awareness for the anthropogenic impact on the environment has greatly increased the past decades.

Human activities have been recognized as the major cause of climate change, biodiversity loss and

disruption of nutrient cycles (Hooper et al. 2005; Cardinale et al. 2012). Public reports of massive

animal mortality after e.g. oil spills such as the Deepwater Horizon disaster (Abbriano et al. 2011) are

typically observed following accidental discharges of chemical waste e.g. the 2015 dam burst in a

Brazilian mine (UN Human Rigths 2015). However, pollution also affects ecosystems at smaller spatial

scales and many effects are more subtle and do not necessarily lead to mass mortality of individuals

(Fleeger et al. 2003). In order to prevent such effects, it is important to quantify the risk a chemical poses

to the ecosystem to make informed decisions about its production and use. Most chemicals are potential

hazards for ecosystems i.e. they have inherent properties that can cause damage to the ecosystem (van

Leeuwen and Vermeire 2007). The ecological risk of a chemical refers to the probability that a chemical

will cause damage to the ecosystem, taking into account exposure. Quantification of the risk a chemical

poses for the ecosystem is done by performing an ecological risk assessment (ERA).

In general, the goal of ecological risk assessment of chemicals is to quantify the risk that a concentration

of a given chemical would impair the structure and functioning of natural ecosystems and to derive

maximum environmental concentrations that prevent ecological effects (Preston 2002; De Laender and

Janssen 2013). Typically, ecological risk assessment is divided in two parts: exposure assessment and

assessment of the potential ecological effects (Figure 1.1; van Leeuwen and Vermeire 2007). Exposure

assessment is used to determine the concentration to which the ecosystem will be exposed. In Europe,

this is typically the Predicted Environmental Concentration (PEC). Chemicals in the environment

undergo different processes such as (bio)degradation, absorption and evaporation (Chapman et al. 1998).

All these processes will determine the final bioavailable concentration i.e. the concentration to which

organisms are actually exposed. Effect assessment is used to determine the Predicted No Effect

Concentration (PNEC), a threshold environmental concentration below which effects to the ecosystem

are not expected to occur. Traditionally, the PEC value is divided by the PNEC to calculate the Risk

Quotient (RQ). RQ values higher than 1 indicate potential risk when the chemical of concern would be

released in the environment without mitigation measures.

General introduction and outline

3

Figure 1.1: Scheme of a traditional ecological risk assessment. Adapted from van Leeuwen and Vermeire 2007.

Box 1.1: Selection of important environmental legislation in the European Union and their respective environmental objectives.

Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH; EC 2007): to regulate

the production and use of chemicals.

Protection of human health and the environment

Increased transparency

Promotion of non-animal testing

Integration with international efforts

Water Framework Directive (WFD; EC 2000; SCHER et al. 2013): to protect the ground and surface

waters of Europe against environmental pollution

Good ecological status of aquatic water bodies. In other words, the characteristics of aquatic

ecosystems should be as close as possible to the reference conditions of natural water bodies

not subject to human pressure.

Regulation 91/ 414/EEC: to regulate plant protection products (EC 2004)

The use of plant protection products does not have any long-term repercussions for the

abundance and diversity of non-target species. This implies protection of populations and

communities (not individual organisms), and the possibility of accepting some short-term

effects if followed by recovery.

Chapter 1

4

Legislation has been developed to prevent environmental risk by performing predictive risk assessments

and specifying formal environmental protection goals. In Europe, several legislations were adopted over

the years (Box 1.1). Biodiversity and ecosystem functions are usually specified as protection goals

(Hommen et al. 2010). For example, the European Commission has made a commitment towards

“halting the loss of biodiversity and the degradation of ecosystem services in the EU by 2020” (EU,

2010) and, according to the Water Framework Directive (WFD), all European surface and groundwater

bodies should have a good ecological status by 2015 (EU, 2000). The focus on biodiversity in legislation

is understandable as biodiversity is generally considered an useful descriptor of ecosystem structure and

its role in ecosystem productivity and stability is generally accepted in the ecological literature (Hooper

et al. 2005, 2012; Cardinale et al. 2012).

1.3. Current ERA methods and their limitations

The number of registered chemicals is approximately 100,000 and still increasing (Clements and Rohr

2009). To cope with this high number of chemicals, ERA is typically conducted using a tiered approach

(Figure 1.2) i.e. chemicals that pose a higher risk at lower tiers are subjected to more extensive and

complex risk assessment methods at higher tiers (Brock et al. 2006; SCHER et al. 2013). The lowest tier

is primarily used for screening chemicals i.e. identification of chemicals that could pose a risk. This

lowest tier is very similar in all EU directives and is based on the risk quotient i.e. the ratio of the PEC

and the concentration at which effects are expected (Hommen et al. 2010). In all directives except for

the plant protection directive, the Predicted No Effect Concentration (PNEC) is used as the reference

concentration for effects. The PNEC is calculated by dividing the endpoint of the most sensitive test

species by an appropriate assessment factor. Assessment factors, also sometimes called safety factors,

are used to account for uncertainty concerning the accuracy of the selected endpoint such as intra- and

interspecies differences in sensitivity, differences between acute and chronic tests and the lab-to-field

extrapolation (Chapman et al. 1998). If the risk quotient is larger than 1, the chemical has a potential

risk to the environment and needs to either undergo higher tier risk assessment to further assess the risk

or be risk managed.

The use of the risk quotient, and especially assessment factors, has been heavily criticized in ERA.

Assessment factors are a conservative method to deal with the uncertainty related to the extrapolation

to real situations but are based on policy and not on science (Chapman et al. 1998; Forbes et al. 2008).

Indeed, this approach relies too much on expert judgement to relate risk ratios to environmental

protection goals (Forbes et al. 2009a). This often leads to an overestimation of the risk and consequently,

unrealistically low “safe” concentrations (Chapman et al. 1998). Also, this approach only uses the most

sensitive endpoint of the available toxicity tests (Forbes et al. 2008). Unless additional toxicity tests

reduce the assessment factor e.g. a chronic test versus an acute test, they are only used to calculate a


5

new PNEC when the measured endpoint is even more sensitive than the previous one. More information

on the potential toxicity of a chemical does thus not necessarily lead to more accurate risk assessments

with this approach.

Figure 1.2: The tiered approach to the risk assessment of chemicals. Chemicals that pose a higher risk may be subject to more extensive assessment methods.

Higher tier risk assessment methods include the use of extrapolation models and multi-species test

systems. The species sensitivity distribution (SSD) is the most used extrapolation model and fits a

probability distribution to a set of toxicity thresholds derived in single species toxicity tests (Forbes and

Calow 2002; Posthuma et al. 2002). The probability distribution is used to determine a concentration at

which a certain percentage of the species is not affected. Typically, the concentration at which 95% of

the species are protected is used as the PNEC value. A major advantage of this approach compared to

the risk quotient technique is that it incorporates all information available from different species. The

SSD approach has been compared with model ecosystem data and field data for many chemicals e.g. for

endosulfan (Hose and Van den Brink 2004) and fluazinam (van Wijngaarden et al. 2005). Indeed, most

SSD-derived treshold concentrations were protective for ecosystem structure and functioning (Versteeg

et al. 1999). However, the properties and underlying assumptions of SSDs have been discussed in depth

over the years and questions, mainly related to their underlying assumptions, have been raised about

their use for ecological risk assessment (Forbes and Calow 2002).

One of the main issues is that the species included in the SSD are considered as components of a realistic

community while this is rarely the case (Forbes and Calow 2002). Generally, any organism for which

sensitivity data are available is included and this species is assumed to be equally important for structure

and functioning of the ecosystem. This ignores the possible presence of keystone species which have a

larger than average contribution to ecosystem structure and functioning. Also, SSDs assume that

interactions between individuals and species will not influence the sensitivity of the community (De

Laender et al. 2008c; Schmitt-Jansen et al. 2008). However, there are many examples of how ecological

interactions can influence the outcome of chemical exposure and this assumption is thus unrealistic. For

example, one modelling study compared a SSD approach that neglected ecological interactions with one

that accounted for ecological interactions (De Laender et al. 2008c). The latter study showed that for

approximately 25% of the toxicants, the SSD approach that took ecological interactions into account

Chapter 1

6

was more strict than the SSD approach that neglected ecological interactions. Therefore, it is impossible

to determine if derived safe concentrations with a SSD are protective for real ecological communities.

Several statistical considerations need to be accounted for when applying SSDs. To calculate accurate

percentiles, a sufficient number of species needs to be included and the appropriate distribution should

be used. The amount of species needed differs between cases but has been estimated to range between

15 and 55 species (Newman et al. 2000). This is higher than what is required for most regulatory

purposes and for most chemicals this amount of data is not available. Also, the identity of the species

included in the SSD will influence the derived safe concentrations (Forbes et al. 2001; De Laender et al.

2013). Inclusion of a large number of species sensitive to a certain chemical e.g. primary producer for a

photo-synthesis inhibiting herbicide, will lead to lower safe concentrations than a set of species where

primary producers are under-represented (Van den Brink et al. 2006). To fit the SSD to the data, the log-

normal distribution is often chosen but this distribution is often not applicable to the data (Newman et

al. 2000). These statistical considerations are often neglected which leads to inaccurate predictions and

a high uncertainty on the derived percentiles and thus ‘safe’ concentrations (Forbes and Calow 2002).

Finally, questions can be raised about how a SSD is used and interpreted. Often the HC5 i.e. the

concentration at which 5% of all species are affected, is calculated and used as a safe environmental

concentration. This assumes that 5% of the species is an appropriate protection level i.e. that the loss of

5% of the species does not affect the ecosystem structure and functioning (functional redundancy) and

that biodiversity, as a measure of ecosystem structure, is a more sensitive ecosystem endpoint than

ecosystem functions (Newman et al. 2000; Forbes and Calow 2002; De Laender et al. 2008a). This

appears to be true for herbicides, insecticides and fungicides: comparison between SSD-derived HC5

values and no-effect concentrations in model communities showed that HC5 values were, in general,

protective for at least short-term exposure (Van den Brink et al. 2006; Maltby et al. 2009). For

insecticides, the SSD-approach was protective in 25 of the 27 cases when compared with experiments

with model communities (van Wijngaarden et al. 2015). Further evaluation is however required for

chemicals that have chronic toxicity or modes of action that have been rarely tested e.g. neonicotinoids.

Experiments with model communities, both small scale (microcosms) and large scale (mesocosms), are

currently considered the most ecologically relevant effect assessment techniques because they expose

realistic aquatic communities over a longer period of time to the chemical (Schmitt-Jansen et al. 2008).

This approach can account for ecological interactions and indirect effects i.e. effects on tolerant species

through interactions with sensitive species (Fleeger et al. 2003; De Laender et al. 2011). Population and

ecosystem recovery can also be assessed with this method. However, micro- and mesocosms also have

several disadvantages. The amount of time and resources required to perform such experiments is a

major drawback (De Laender et al. 2013). Also, interpretation of the results is non-trivial, although

methods like the principal response curve technique have been developed to deal with this (Van den

Brink and Ter Braak 1998). Other disadvantages include problems with scaling effects, the protection


7

of rare species, sensitivity to starting conditions and the inability to replicate every natural system

(Forbes et al. 2008; Schmitt-Jansen et al. 2008). Effects occurring in the model communities do not

necessarily correspond with effects at more realistic spatial scales, referred to as scaling effects (Forbes

et al. 2008). For example, in realistic landscapes, migration of individuals can alter the observed effects.

Moreover, model communities might miss effects on rare species not present in the sampled community.

Species can also be sensitive to the starting conditions of the model community experiment, reducing

the relevance of the experiments (Hjorth et al. 2007). Lastly, it is impossible to perform such

experiments for each natural system. Comparison between different communities exposed to the same

chemical has shown that, although the sensitive species are affected at similar concentrations, indirect

effects and recovery can indeed be very different (Daam and Van den Brink 2010).

In general, all current ERA methods fail to provide an accurate and certain answer to the central question

in ecotoxicology: what are the large-scale effects of chemical stress in real-world systems (Beketov and

Liess 2012). As a result, decisions in risk assessment are accompanied with large uncertainty and the

prescribed protective concentrations are possibly either under- or overprotective. The major problem is

how to extrapolate from these, at best, simple community tests performed in a controlled environment

to the protection goals set by the authorities (De Laender et al. 2008a; Forbes et al. 2008). The

relationship between typical ecotoxicological endpoints such as growth, survival and fecundity and

population or ecosystem dynamics is complex, non-linear and thus difficult to predict using simple

techniques (Forbes et al. 2008). Therefore, ecological risk cannot be adequately assessed using

procedures that disregard most of the inherent environmental and ecological complexity (De Laender et

al. 2014a). However, current ERA procedures fail to consider newly developed methods such as

ecological models that were specifically developed to reduce the uncertainty in ERA. Also, it is unclear

that current approaches are able to accurately predict future risks, especially considering that the number

of environmental stressors are increasing (Grimm and Martin 2013). Multiple stressors can refer to a

combination of different chemicals but also to the combination of chemicals and other abiotic stressors

such as temperature (De Laender and Janssen 2013). How these multiple stressors interact is difficult to

predict (Gabsi et al. 2014b) but it is clear that the presence of multiple stressors may have potentially

large implications for ERA.

1.4. Integration of ecology into the risk assessment procedure

In order to more accurately predict the effects of chemicals on communities and ecosystems, more

ecology needs to be integrated in ERA approaches (Chapman 2002; Clements and Rohr 2009; Grimm

et al. 2009). The call for integration of ecological processes was already brought up as early as the 1980s

– “putting the eco in ecotoxicology” (Cairns 1988) – and 1990s (Baird et al. 1996). Recent notable

efforts call for the integration of macro-ecology in ecotoxicology (Beketov and Liess 2012) and for the

Chapter 1

8

application of community ecology to ecotoxicological theory (Schmitt-Jansen et al. 2008). Scientific

efforts have been done to address these concerns e.g. the use of ecological models (Galic et al. 2010).

However, integration of more ecology at the regulatory level has thus far been limited to the use of

micro- and mesocosm experiments as highest tier ERA tools. Criticism of the current ERA procedures

was summarized in an opinion report of different Scientific Committees of the European Union (SCHER

et al. 2013). The report recognizes that current ERA procedures lack ecological realism which leads to

high uncertainty associated with the predictions made.

With the current advances in ecological modelling and informatics, it should be feasible to use more

complex and computationally intensive techniques that are better at integrating ecological principles.

Several ecological features have been identified or suggested as important to consider in ERA. These

include, but are not limited to, ecological interactions, patterns in exposure, the spatial structure and

scale, the presence of keystone species, functional redundancy and recovery potential (Figure 1.3). Three

of these are addressed in detail in this PhD thesis: ecological interactions, spatial scale and structure and

recovery. Keystone species are not the focus of this PhD because they are not present in every

community. Effects on functional redundancy is related to effects on ecosystem functioning while this

work focuses more on patterns in species abundances. However, the tools developed in this thesis could

easily be adapted to cover these two cases e.g. by adjusting the community to incorporate a keystone

species or by monitoring ecosystem functions such as primary or secondary production.

Figure 1.3: Overview of different ecological aspects that are often overlooked with traditional ecological risk assessment approaches. Aspects covered in this PhD thesis are marked in green.

In ecosystems, individuals exposed to a chemical are not isolated but interact with individuals of the

same and/or of other species. However, traditional ERA approaches regard individuals as discrete units

instead of interacting entities (Preston 2002). In more diverse communities the number and complexity

of interactions is higher than in less diverse communities (Relyea and Hoverman 2006). Accurately

assessing species interactions is essential to perform ecologically realistic risk assessments (De Laender

et al. 2014a). In natural systems, indirect effects of chemicals - effects on more tolerant species through


9

interactions with sensitive species - are common (Rohr et al. 2006; Clements and Rohr 2009). Because

they involve multiple species, indirect effects are also intrinsically more complex and more difficult to

predict (Rohr et al. 2006). Competition and predation are regarded as the most important ecological

interactions when considering indirect effects of chemicals (Preston 2002).

Competition can refer to competition for the same food source but also to competition for space, light

or other limiting resources. Competition can occur between individuals of different species (interspecific

competition) but also within one population of the same species (intraspecific competition). Generally,

when chemical stress leads to decreased population densities, the surviving individuals in the population

experience less intraspecific competition (Foit et al. 2012). This decreased intraspecific competition

reduces the effect of chemical stress on population densities (Liess 2002; Foit et al. 2012; Del Arco et

al. 2015) and allows more rapid recovery of population density or size structure after exposure (Foit et

al. 2012; Knillmann et al. 2012b). For competition between tolerant and sensitive species, tolerant

species are expected to be, at least partly, relieved from competition (Foit et al. 2012). Consequently,

the tolerant species perform better after exposure to stress. This was experimentally shown for Daphnia

magna and Culex sp. larvae exposed to fenvalerate (Foit et al. 2012) and for Daphnia spp. in microcosms

exposed to esfenvalerate (Knillmann et al. 2012b). Other examples include modelling studies where

competition prolonged the effects of the chemical (Kattwinkel and Liess 2013) or increased the

vulnerability of the population to chemical stress (Gabsi et al. 2014b). In reality, species can interact in

other ways than only via competition. For example, in an experiment with Asellus aquaticus and

Gammarus pulex, competition positively influenced G. pulex survival during exposure to carbendazim

(Del Arco et al. 2015). This was attributed to predatory compensatory mechanisms by G. pulex under

food-limited conditions, showing that reality can be much more complex and unpredictable than

standard experiments would suggest.

For predation, chemical stress can either affect the prey, the predator, or both. Similarly to interspecific

competition, effects on the predator may relieve the prey species from predation, allowing it to increase

in abundance (Fleeger et al. 2003). For example, phytoplankton species increased in abundance after

elimination of the grazers due to carbendazim toxicity (Van den Brink et al. 2000). A common

observation when prey are more sensitive than predators is that prey exposed to a toxicant are more

vulnerable to predation (Fleeger et al. 2003; Beketov and Liess 2006). For example, Artemia sp.

populations went extinct after combined exposure to chemical stress and simulated predation (Beketov

and Liess 2006). In this case, the predation pressure prevented density-dependent compensation for the

chemical effect i.e. increased reproduction at low densities was not possible. Chironomus larvae were

less active after cadmium exposure (Rooks et al. 2009). This increased their susceptibility to active

predators but interestingly, the predation rate by ambush predators was not affected. Predator species

can also be affected by chemical stress through their prey e.g. ciliates starved when their food source

was eliminated by prometryn (Liebig et al. 2008). However, predation does not always lead to higher

Chapter 1

10

effects of a chemical. Presence of predator kairomones resulted in Daphnia magna producing larger

offspring that were more resistant to chemicals, thus reducing the effect of carbaryl exposure (Coors and

Meester 2008; Gergs et al. 2013). How chemical stress interacts with competition and predation is

clearly chemical and species specific. To predict how these interactions influence the effect of

chemicals, a better understanding of the underlying processes is needed.

Differences in application time, emission rates and location can result in significant differences in timing

and levels of exposure, ultimately leading to different effects on local populations (Galic et al. 2012).

Depending on the chemical identity, exposure patterns can differ greatly (De Laender et al. 2014a).

Chemicals differ in their application time, partition to different compartments in the environment and

differ in their persistence in the environment. These realistic exposure patterns can differ greatly from

typical lab tests with short, constant exposure. Another important aspect is the timing of exposure in the

life cycle of the exposed populations. Early life stages, especially embryonic stages, are often more

sensitive to chemicals. Exposure during periods of reproduction can thus lead to larger population effects

than exposure during periods of no reproduction, especially for species with long life cycles (Bridges

2000; Galic et al. 2012). The landscape structure and the presence of unexposed populations is another

factor determining the outcome of chemical exposure and is especially important for the recovery of the

affected populations. For example, isolated communities recovered more slowly from the application of

endosulfan than less isolated communities (Trekels et al. 2011). Similarly, the presence of

uncontaminated patches decreased the recovery time of mesocosm communities after lufenuron

exposure (Brock et al. 2009).

“Keystone species” are species that are essential for certain ecosystem functions or that enable other

species to survive in the ecosystem (Chapman 2002). Keystone species often indicate the presence of

an ecological threshold and the loss of these keystone species often results in abrupt, non-linear changes

in ecosystem structure and functioning that are difficult to recover from (Clements and Rohr 2009). For

example, the burrowing ghost shrimp Neotrypaea californiensis is an important facilitator for many

other species of the soft sediment benthos. Direct effects of carbaryl on this shrimp species also

indirectly affected the associated benthos species (Dumbauld et al. 2001). Therefore, it is important to

identify the presence and identity of keystone species in ecosystems. Closely related to this is the concept

of functional redundancy. Functional redundancy occurs when the loss of some species does not result

in the loss of ecosystem functioning because more tolerant species compensate for the affected species

(Chapman 2002). Not all species are thus always equally important for the ecosystem structure and

functioning and ecological risk assessments should take this into account.

Another often neglected characteristic of natural populations is their potential to recover after a

disturbance event. In natural ecosystems, populations are regularly disturbed by environmental factors.

If we consider the long-term effects of chemicals, the recovery potential of a population or ecosystem


11

is thus an important factor (Relyea and Hoverman 2006; Clements and Rohr 2009). The recovery

potential differs between species and systems e.g. short lived species will typically recover quicker from

chemical stress than species with a longer life cycle, and is influenced by other environmental factors

e.g. indirect effects (De Lange et al. 2010; Knillmann et al. 2012b). Ideally, ecological risk assessments

should take the recovery potential of a system into account and be more stringent when the recovery

potential is low.

1.5. Modelling as a risk assessment tool

A main challenge in the field of ecotoxicology is to develop tools that can take into account the

ecological complexity displayed in real ecosystems (Figure 1.3) so that site-specific effects can be

assessed. Ecological modelling was proposed as one of the best options to improve effect assessment,

specifically to account for ecological interactions and spatial and temporal variability in exposure

(SCHER et al. 2013). However, the use of models is far less accepted in effect assessment than in

exposure assessment. In particular, the development of FOCUS (FOrum for the Co-ordination of

pesticide fate models and their Use; FOCUS 2001) models and scenarios provided a standardized way

to develop and use environmental fate models (Grimm and Martin 2013; SCHER et al. 2013). The

inherent complexity of ecosystems, the apparent lack of universal ecological laws and the lack of clear

protection goals have hindered the acceptance of models as tools for effect assessment of chemicals

(Van Straalen 2003; Van den Brink et al. 2006). In legislation, ecological models are thus mostly ignored

as possible ERA tools. Only the guidance document relating to aquatic toxicology under Directive

91/414/ EEC (SANCO 2002) lists ecological models as possible tools in higher tiered risk assessments

for extrapolation from microcosm or mesocosm studies to the field (Hommen et al. 2010). It is clear that

more ecological knowledge and more complex decision making is required to apply ecological models

as tools for effect assessment (Dohmen et al. 2015). Indeed, modelling could be an ecology based

alternative to the standard ERA approaches and can actually be used to assess the effects of chemicals

on the actual protection goals i.e. biodiversity and ecosystem functioning (De Laender et al. 2008b;

Forbes et al. 2009a). Potential applications of models in ERA include (i) evaluating the relevance of

effects on individuals for population dynamics, (ii) extrapolation to untested exposure patterns, (iii)

extrapolation of recovery processes from lab to field, (iv) assessment of indirect effects and (v)

evaluation of bioaccumulation and biomagnification of chemicals (Hommen et al. 2010). Several

initiatives have been taken during the past decade to explore and promote the use of ecological models

in risk assessment e.g. the CREAM (Grimm et al. 2009) and ChimERA (De Laender et al. 2014a)

projects, as well as through the establishment of the MemoRisk SETAC advisory group (Preuss et al.

2009b) and the organization of several workshops e.g. LEMTOX (Forbes et al. 2009b) and

MODELLINK (Hommen et al. 2016).

Chapter 1

12

Using models to assess the effects of chemicals on populations, communities or ecosystems has several

advantages. Most importantly, models can help clarify how chemicals affect higher levels of biological

organization, why certain effects are happening and what the most important drivers are (Grimm et al.

2009). Ecological models have successfully been used to comprehend and predict the effects of

chemicals on populations (Galic et al. 2010; Preuss et al. 2010; Dohmen et al. 2015), communities (De

Laender et al. 2014b), and ecosystems (De Laender et al. 2015). With advances in computational power

and efficiency, the explicit consideration of time and space in models is becoming more feasible (Galic

et al. 2010). This allows the modelling of populations in heterogeneous landscapes, where the exposure

can be drastically different between different locations and different times of the year. Modelled Asellus

aquaticus populations were predicted to recover faster when connectivity in the habitat was higher

(Galic et al. 2012). Exposure during periods of reproduction resulted in slower recovery, indicating the

importance of the exposure profile. Models are also ideally suited to study how effects on individuals

translate to effects on population and ecosystem dynamics (Bradbury et al. 2004; Forbes and Calow

2012). For example, evaluation of different physiological modes of actions in Daphnia magna

populations showed that direct effects on survival and reproduction had a much larger impact on

population densities than effects on growth or feeding rate (Gabsi et al. 2014a). Modelling can thus help

reduce the uncertainty associated with the extrapolation from ecotoxicological observations to

ecological effects (Forbes et al. 2008).

Modelling approaches can also be informative for the lower tiers of risk assessment. Experiments in

silico can help to design toxicity tests, interpret individual responses to chemical stress and identify

deficits in current datasets, allowing for more focused experimental work (Galic et al. 2010; Jager et al.

2014). Moreover, the large amount of historical ecotoxicology data can be used for modelling purposes

and models can reduce the need for additional ecotoxicological tests, reducing the amount of test animals

needed.

1.6. Individual based modelling

Individual based models (IBMs) seem particularly suited for use in ERA. IBMs consider processes

occurring at the individual level such as feeding, growth and reproduction (Martin et al. 2013b; Gabsi

et al. 2014b). Population properties are not modelled directly but emerge from the individuals in the

population. Since most ecotoxicological tests focus on the individual level, IBMs are ideal tools to

translate these test results to the population level. Incorporating chemical effects on individuals in IBMs

allows exploration of the effects of chemicals at the population level. In recent years, IBMs have been

used to predict the population dynamics of a number of typical test species used in ecotoxicology e.g.

Daphnia magna (Preuss et al. 2010; Martin et al. 2013a) and Asellus aquaticus (Galic et al. 2012).

Similarly, the effects of a hypothetical insecticide on three populations of arthropods were modelled


13

using individual based models (Dohmen et al. 2015). These IBM applications neglected possible

interactions with other species and focused solely on a single species. However, the absence of

interactions between species is one of the main criticisms on current ERA methods (Rohr et al. 2006;

Clements and Rohr 2009). Considering that ecological models have been suggested as tools to

extrapolate individual-level effects observed in experiments to population, food web and ecosystem

level effects (Grimm et al. 2009; De Laender et al. 2014a), it is important to develop ecological models

that do incorporate interactions between species. One recent modelling effort showed that adding

interspecific competition to individual-based models increased recovery times following chemical stress

up to three times (Kattwinkel and Liess 2013). IBMs have however not been applied yet for more

complex systems e.g. food webs, where more species and more interactions between these species need

to be considered.

1.7. Current limitations to using modelling in risk assessment

Modelling could be useful to overcome many of the shortcomings of current ERA procedures. However,

modelling needs to improve in certain key areas before it will be fully accepted as an ERA tool. These

improvements are two-fold: (1) improvements of the science underlying the models and (2)

improvements of the regulatory use of models (Grimm and Martin 2013). To improve the science behind

the models, validation is of key importance i.e. how can we know that developed models capture reality,

that the science is sound (Forbes et al. 2008; Grimm and Martin 2013)? Validation is, however, a very

broad term and unclear terminology is often an obstacle to understand what validation entails (Augusiak

et al. 2014). A better defined validation process such as the evaludation process defined by Augusiak et

al. 2014, will increase the acceptance of models and increase the understanding of the science behind it.

Validation is closely linked to the emergence of patterns from data (Grimm and Martin 2013).

Observations are considered patterns if they are unlikely to result from random processes and thus

contain information about the underlying organization. By comparing model output with these patterns,

we can validate the model i.e. assess if the model accurately captures the underlying processes and is

thus a realistic, scientifically sound representation of the world. A related scientific question is how

general does a model need to be, how much complexity does it need, to accurately answer the questions

posed (Forbes et al. 2008). In general, more complex models can predict the outcome more accurately

but are harder to validate. The selection of the correct level of complexity for the focus situation is thus

essential.

To improve the regulatory use of models, risk assessors need to be convinced of their accuracy and

potential (Forbes et al. 2009b; Grimm and Martin 2013). Risk assessors are usually not trained in

modelling and cannot be expected to accurately assess the appropriateness and validity of a model. For

exposure assessment, this problem was solved by developing the FOCUS framework (FOCUS 2001).

Chapter 1

14

This provided risk assessors with a standard way to evaluate exposure models. A similar standardization

is needed for ecological models. Some efforts have been made to promote the use of ecological models

for ERA e.g. The EU-funded CREAM project (Chemical Risk Effects Assessment models; Grimm et

al. 2009; Grimm and Thorbek 2014). Similarly, good modelling practices such as the ODD (overview,

design concepts and details) protocol and TRACE (transparent and comprehensive ecological

modelling) have been developed (Grimm et al. 2006; Schmolke et al. 2010).

1.8. Problem formulation and objectives of this PhD thesis

It is clear that ERA needs state-of-the-art tools to achieve its goal: accurately assessing the risk a

chemical poses to the environment, taking into account essential ecological characteristics of real

systems. One key characteristic that needs to be accounted for are interactions between individuals, both

within one species and between different species. Individual based models (IBMs) are one of the most

promising tools for ERA but their applicability to food webs is unclear. Also, realistic exposure

scenarios, with spatial and temporal variability in exposure, need to be considered. Therefore, the

objectives of this PhD thesis are:

understand how species interactions (competition and predation) interfere with chemical stress;

develop and apply IBMs for two species competing for the same food source;

develop an individual-based food-web model;

develop and apply an integrated ecological risk assessment model (ChimERA) to different

environmental scenarios;

This research is presented in five main chapters (Figure 1.4), increasing the complexity from simple

food webs in the laboratory to more complex food webs in an integrated exposure and effect model. The

main conclusions were summarized in a final chapter.


15

1.9. Thesis outline

1.9.1. Species interactions and chemical stress

Interactions between individuals of the same species and of different species will alter how these

individuals respond to chemical stress. To understand how intra- and interspecific competition and

predation can alter the effect of chemical stress, experiments were performed with simple food webs

(Chapter 2). Communities of the water flea Daphnia magna, the rotifer Brachionus calyciflorus and

the phantom midge Chaoborus obscuripes larvae were exposed to pyrene. Effects of pyrene were

expected to be higher in D. magna populations exposed to strong competition or predation. The tolerant

competitor B. calyciflorus was expected to increase in abundance after pyrene exposure. An indirect

negative effect on the predator C. obscuripes was expected through direct pyrene effects on its prey D.

magna.

1.9.2. Modelling competing species under chemical stress

Models can help to understand and predict the effects of chemicals on populations and food webs. IBMs

in particular are useful because effects of chemicals can be included at the level of the individual, the

focal level of standard ecotoxicity tests. To make the IBMs as widely applicable as possible, they should

be based on a generic theory. In Chapter 3, I describe the development of the DEBkiss IBM: an

individual based model based on the DEBkiss theory. Chemical effects were included in the model by

considering effects on survival using either concentration-response curves or toxicokinetic-

toxicodynamics models.

The validity of the DEBkiss IBM framework was tested in Chapter 4 by applying IBMs to the

experiments with D. magna and B. calyciflorus described in Chapter 2. More specifically, I evaluated

whether the DEBkiss IBM framework could simulate the population densities of these two grazers when

exposed to pyrene, interspecific competition and both pyrene exposure and interspecific competition.

This was done by comparing the model simulations with the patterns observed in the experiment.

1.9.3. Food web model development and integration with a fate model: the ChimERA

model

Realistic ecosystems are not limited to two species but consist of many interacting species. In Chapter

5, predation was therefore added to the competition implementation presented in Chapters 3 and 4 and

used to develop an individual-based food web model. Although traditionally separated, exposure and

effect assessment are both integral parts of ecological risk assessment. Chapter 5 is therefore concluded

Chapter 1

16

with the integration of a fate model with the food web model, forming the ChimERA model: a spatially-

explicit model simulating both the fate of a chemical and the effects on the food web.

1.9.4. Scenario analysis with the ChimERA model

A key feature of the integrated ChimERA model is that the effects of different environmental parameters

on the food web dynamics can be evaluated, opening a whole new, prospective approach to ERA. In

Chapter 6, the influence of trophic state, temperature, hydrodynamics and chemical exposure pattern

on the resulting chemical effects were evaluated. Effects of these environmental variables on the

environmental fate of the chemical and on the food web dynamics was assessed.

1.9.5. Conclusion and future directions

In the final Chapter 7, the findings of this dissertation are reviewed and summarized in a set of

conclusions. Suggestions and possible directions for future research are provided.


17

Figure 1.4: Schematic overview of the chapters included in this dissertation and the subjects they discuss.

Chapter 11 Chapter 111 & IV

...... • ..... Competition

Experiment lndividual based models

Lab exposure Lab exposure

Predation

ChapterVI ChapterV

Scenario analysis

Sce~ario 11 Sce;ario 11 Sce;ario

Chemica! fate model

19

2 SPECIES INTERACTIONS AND CHEMICAL STRESS:

COMBINED EFFECTS OF INTRASPECIFIC AND

INTERSPECIFIC INTERACTIONS AND PYRENE ON

DAPHNIA MAGNA POPULATION DYNAMICS

Redrafted from:

Viaene KPJ, De Laender F, Rico A, Van den Brink PJ, Di Guardo A, Morselli M, et al. Species

interactions and chemical stress: Combined effects of intraspecific and interspecific interactions and

pyrene on Daphnia magna population dynamics. Environ Toxicol Chem. 2015;34(8):1751–9.

Chapter 2

20

Abstract

Species interactions are often suggested as an important factor when assessing the effects of chemicals

on higher levels of biological organization. Nevertheless, the contribution of intra- and interspecific

interactions to chemical effects on populations is often overlooked. In the current chapter, Daphnia

magna populations were initiated with different levels of intraspecific competition, interspecific

competition and predation and exposed to pyrene pulses. Generalized linear models were used to test

which of these factors significantly explained population size and structure at different time points.

Pyrene had a negative effect on total population densities, with effects being more pronounced on

smaller D. magna individuals. Among all species interactions tested, predation had the largest negative

effect on population densities. Predation and high initial intraspecific competition were shown to

interact antagonistically with pyrene exposure. This was attributed to differences in population

structure prior to pyrene exposure and pyrene-induced reductions in predation pressure by Chaoborus

sp. larvae. The results presented in this chapter provide empirical evidence that species interactions

within and between populations can alter the response of aquatic populations to chemical exposure. It

is concluded that such interactions are important factors to be considered in ecological risk

assessments.

21

2.1. Introduction

Current procedures for the ecological risk assessment (ERA) of chemicals are generally based on the

extrapolation of individual-level responses to the whole ecosystem and often fail to integrate a sufficient

level of ecological realism (De Laender et al. 2008b; SCHER (Scientific Committee on Health and

Environmental Risks) et al. 2013). In ecosystems, individuals exposed to a chemical are rarely isolated

but interact with individuals of the same and/or of another species. Despite being one of the key

characteristics of ecosystems, interactions within and between species are rarely included in current

prospective ERAs, especially for non-pesticidal chemicals (De Laender et al. 2008c). However, species

interactions can alter the direct effects of a chemical on a sensitive species e.g. increased mortality after

pesticide addition because of decreased predator avoidance behavior (Hanazato 2001). Alternatively, by

interacting with sensitive species, tolerant species can also be affected leading to indirect effects of

chemical stress e.g. starvation of the consumer species when the prey species is affected (Rohr and

Crumrine 2005; De Hoop et al. 2013) or reduced competition with the affected species (Rohr and

Crumrine 2005). The indirect effects of a chemical are often overlooked but can be as large or even

larger than the direct effects of the chemical (Fleeger et al. 2003). Interactions with other species can

either increase or decrease the susceptibility of populations and communities to a chemical (Preston

2002; Fleeger et al. 2003). For example, the no observed effect concentration (NOEC) of prometryn for

ciliates was more than two orders of magnitude lower in microcosms compared with a single-species

toxicity test because of the sensitivity of their food source to prometryn (Liebig et al. 2008). Also,

elimination of grazers by the fungicide carbendazim allowed certain phytoplankton species to increase

in abundance (Van den Brink et al. 2000) and exposure to insecticides resulted in the development of

anti-predator structures in daphnids, potentially reducing the effect of predation (Hanazato 2001).

Accurately assessing species interactions is thus essential to perform ecologically realistic chemical risk

assessments (De Laender et al. 2014a).

Competition and predation are regarded as the most important species interactions when considering

indirect effects of chemicals (Preston 2002). Competition can occur between individuals of different

species (interspecific competition) but also within one population of the same species (intraspecific

competition). Although several studies exist on the combined effects of interspecific competition and

chemicals (Foit et al. 2012; Knillmann et al. 2012b), studies on how intraspecific competition affects

the response of populations to chemical exposure are rather underrepresented in the ecotoxicological

literature.

The objective of the current chapter was to investigate how initial differences in species interactions

influence the response of aquatic invertebrate populations to chemical stress. To this end, Daphnia

magna populations were initiated with different levels of intraspecific and interspecific competition and

Chapter 2

22

predation with pyrene as a chemical stressor. Population size and structure of D. magna were evaluated

using generalized linear models. Higher effects of pyrene were expected in populations that are

experiencing increased competition or predation pressure compared to a control population.

2.2. Materials and Methods

2.2.1. Experimental design

D. magna populations were exposed to six levels of species interactions (i.e. species interaction control,

low and high intraspecific competition, low and high interspecific competition, and predation) and to

five different pyrene exposure profiles (i.e. control, solvent control, and low, medium and high exposure;

see Table 2.1). The experiment was performed in triplicate (n = 3). Two additional replicates were added

for the species interaction control treatment without pyrene exposure (n = 5). In a follow-up experiment,

referred to as experiment 2, D. magna populations were exposed to continuous interspecific competition

and to five different pyrene exposure profiles (i.e. control and four different pyrene concentrations; see

Table 2.1). The experiments were carried out in glass vessels (1.5 L) filled with 0.5 L of fresh water RT

medium (Tollrian 1993). The test vessels were randomly distributed within a water bath placed in a

temperature-controlled room (20.8 ± 1 °C) and exposed to low artificial light conditions (1000-1500

lux). Experiment 1 lasted for 29 days with an adaptation period of 7 days (day -7 until day 0). Pyrene

was added twice, on day 0 and day 8. After the second pyrene addition, population densities were

monitored for another 14 days until day 22. In the follow-up experiment, pyrene was added once after

15 days of adaptation (day -15 until day 0) and population densities were monitored for another 16 days

until day 16.

The D. magna organisms used in the experiment were obtained from the laboratory culture of the

department of Aquatic Ecology and Water Quality Management from Wageningen University (The

Netherlands). Scenedesmus obliquus was used as a food source for the D. magna cultures prior to the

experiment and throughout the course of the experiment. Test vessels were fed six times a week with S.

obliquus (1 mg carbon ∙ L-1 ∙ day-1). The rotifer Brachionus calyciflorus, which also feeds on S. obliquus,

is expected to compete with D. magna for food and was used to simulate interspecific competition. B.

calyciflorus cysts were obtained from MicroBioTest Inc.© (Mariakerke, Belgium) and a stock culture

was set up in RT medium at 20°C. Chaoborus sp. larvae, which were added to simulate predation, were

collected from unpolluted mesocosms at ‘de Sinderhoeve’ research station (www.sinderhoeve.org,

Renkum, The Netherlands).

23

Table 2.1. Overview of the different species interactions tested in Experiment 1. The columns indicate how many individuals of each species were added to the test vessels for the different species interaction treatments. Each of these treatments was exposed to five different pyrene exposure profiles in experiment 1: no pyrene, solvent control, low, medium and high pyrene exposure. In experiment 2, each of the treatments was exposed to four different pyrene exposure profiles.

Treatment Number of D. magna

Number of B. calyciflorus

Number of Chaoborus sp. larvae

Experiment 1

Control 10 0 0

Intraspecific competition: low 20 0 0

Intraspecific competition: high 40 0 0

Interspecific competition: low 10 333 0

Interspecific competition: high 10 999 0

Predation 10 0 1

Experiment 2

Control 10 0 0

Competition 10 200 week-1 0

For both experiments, identical D. magna population structures were introduced in all test vessels. They

were composed of 20% adults, 40% juveniles and 40% neonates. The classification of D. magna

organisms within these three groups was based on size, and was performed by filtering the culture

medium through sieves with different mesh sizes (i.e. adults > 800 μm; juveniles between 800 and 500

μm; and neonates < 500 μm) (Preuss et al. 2009a). Neonates typically correspond to individuals younger

than 48 hours. By using populations composed of different life stages, I wanted to simulate realistic

population structures and study the sensitivity of different life stages and its implications for D. magna

population dynamics.

In the first experiment, the effect of intraspecific competition on D. magna populations was studied by

using initial densities of 10 (species interaction control), 20 (low intraspecific competition) and 40 (high

intraspecific competition) D. magna individuals per test vessel. To study how interspecific competition

affected the D. magna population, B. calyciflorus was added to the test vessels at the start of experiment

1 in densities of approximately 333 rotifers ∙ vessel-1 (low interspecific competition) and 999 rotifers ∙

vessel-1 (high interspecific competition). In the follow-up experiment, more continuous competition was

imposed by adding 200 rotifers ∙ vessel-1 weekly. Predation was imposed by the addition of one

Chaoborus sp. larva per test vessel. Chaoborus sp. larvae were added 3 days after the addition of

daphnids to the test vessels to allow the daphnids to acclimatize. When a Chaoborus sp. larva died

during the experiment, it was replaced to assure continuous predation pressure.

Pyrene is a polycyclic aromatic hydrocarbon consisting of four benzene rings. Pyrene was chosen as

model compound for this experiment because of its non-specific, narcotic mode of action (Di Toro et al.

Chapter 2

24

2000). Phototoxicity of pyrene has been reported (Bellas et al. 2008) and experiments were therefore

performed under low light conditions (1000-1500 lux). Acetonitrile was used as solvent for pyrene and,

therefore, a solvent control was included in the experimental design (38 and 75 μg L-1 added for the

first and second addition, respectively). A stock solution of 0.75 g L-1 pyrene was prepared in

acetonitrile and stirred intensively before addition to the test vessels.

In experiment 1, pyrene was applied twice to the test vessels. The first dosing was applied 7 days after

the start of the experiment (day 0) at a nominal concentration of 7.5, 20 and 55 μg L-1 for the low

(Pyrene1A), medium (Pyrene1B) and high (Pyrene1C) pyrene exposure profile, respectively. Pyrene

concentrations were chosen between the EC10 and EC50 values for immobilization. An

EC50,immobilization value of 68 (44-106) μg L-1 was estimated based on a 48 hours toxicity test with

D. magna (OECD 2004) (See Appendix A Figure A1 for the concentration response curve). Using a

similar protocol, no mortality effects were observed for B. calyciflorus and the Chaoborus sp. larvae at

pyrene concentrations up to 150 μg L-1. Because the first pyrene addition had no observable effects on

population densities, pyrene was added a second time at higher concentrations. The second application

was performed 15 days after the start of the experiment (day 8) with a nominal pyrene concentration of

15, 40 and 110 μg L-1, corresponding to the low, medium and high pyrene exposure profile, respectively.

In experiment 2, pyrene was applied only once after 15 days in nominal pyrene concentrations of 20

(Pyrene2A), 50 (Pyrene2B), 100 (Pyrene2C) and 150 μg L-1 (Pyrene2D).

2.2.2. Biological monitoring

In experiment 1, D. magna and B. calyciflorus abundances in the test vessels were monitored on day -

4, 0, 2, 4, 7, 10, 15 and 22. For experiment 2, abundances were counted on day -15, -12, -8, -5, -1, 2, 6,

9, 13 and 16. D. magna were counted and divided into the size classes adult, juvenile and neonate by

filtering the test medium over sieves with mesh sizes of 800 μm, 500 μm and 200 μm, respectively. B.

calyciflorus abundances in the test medium of the interspecific treatments were monitored by taking two

6 mL sub-samples per test vessel and counting swimming rotifers using an inverted microscope

(magnification 10x).

2.2.3. Chemical analyses

Samples for pyrene analysis were taken after the first pyrene application, before the second pyrene

application and two, four and twelve days after the second pyrene application. Pyrene samples were

stored in the dark at -20 °C in glass tubes. The chemical analysis was performed with gas

chromatography–mass spectrometry (Trace GC 2000 series, Thermoquest, DSQ,

Finnigan/Thermoquest). An apolar Zebron ZB 5-ms column (Phenomenex) was used for the analysis,

and extraction and elution were performed by solid-phase extraction according to the manufacturer’s

25

instructions (Waters and Phenomenex). An internal standard (fluoranthene-d10) at a concentration of

10-50 μg L-1 (depending on expected pyrene concentration) was used to control and correct for

extraction losses. The method’s recovery was always >75%. Immediately before injection of the sample,

a recovery standard was also applied to control for the injection itself.

2.2.4. Fate model analysis

The recently developed dynamic water-sediment organism model EcoDyna (Morselli et al. 2014) was

used to predict the temporal fate of pyrene during the experiments. The model was calibrated using the

nominal water volume (500 mL) of the experiment and water-sediment interaction was minimized to

simulate negligible exchange, given the lack of a sediment phase in the vessels used. In order to calculate

potential algal uptake, a daily contribution of 1 mg carbon L-1 was assumed, while organism biomass

was calculated using length-weight relationships (Dumont et al. 1975; Dumont and Balvay 1979).

Physical-chemical properties for pyrene were obtained from literature (Mackay et al. 1992).

2.2.5. Statistical analyses

All analyses were performed using the statistical software package R (version 3.1.1; (R Core Team

2012)). For each sampling time, generalized linear models (GLMs) were constructed. Total, adult,

juvenile and neonate D. magna densities were considered as response variables, allowing for the

examination of population structure. The effect of intraspecific competition (control, low, high),

interspecific competition (control, low, high) and predation (non-predation, and predation) was assessed

for each point in time by constructing a GLM with pyrene and the species interaction considered as the

predictor variables. Time itself was not included as a predictor variable because the effect of time is

non-linear and the effects of the other predictor variables will change over time. The full model is given

by:

The expected density ( ) at time t is the result of the sum of the intercept α, the species

interaction being considered ( ), the pyrene concentration ( ) and their interaction (

).

GLMs were initially constructed assuming a Poisson distribution (Zuur et al. 2009) but this led to

unsatisfactory model validation. I therefore opted to perform GLM analyses with a normal distribution

on the log10-transformed D. magna abundance data. The solvent control treatment was not included in

the GLM analysis as preliminary tests showed no significant differences between the control and the

solvent control treatments. Backwards model selection was used, dropping predictor variables based on

the Akaike’s Information Criterion (AIC), hypothesis testing and model validation analysis (Zuur et al.

Chapter 2

26

2009). As model validation analysis, I (1) inspected if patterns in the data were present using predicted

versus observed plots, (2) inspected if patterns in the residuals were present using predictor versus

residuals plots, and (3) tested the normality of the residuals using QQ-plots (Zuur et al. 2009).

2.3. Results

The effects of the different explanatory variables and their interactions are discussed below. I only

included the results for the total D. magna abundance in this chapter, results for the different size classes

are included in Appendix A.

2.3.1. Pyrene fate

Measured pyrene concentrations in water were lower than expected from the nominal values (Table 2.2).

Nevertheless, there was a clear difference between the three pyrene exposure profiles at any given point

in time. The EcoDyna model was used to simulate pyrene concentration variations in water. The model

was run to fit actual water concentrations, and the importance of the main fluxes dominating the change

in concentration with time after the spikes. As a result of the fitting procedure, it was found that a

chemical half-life in water of 30 h was necessary to reproduce the observed concentrations (no

distinction could be made between biotic and abiotic processes), while volatilization accounted for about

20% of losses. Simulations confirmed that pyrene uptake in algae and animal biomass was negligible.

2.3.2. Population dynamics in absence of pyrene and interactions

Independent of the explanatory variables, a clear trend in the model intercept value could be observed.

In experiment 1, there was an increase in the intercept until day 7, afterwards the intercept slowly

decreased (Table 2.3-2.5). Similarly, in experiment 2 there was a strong population increase the first 14

days, followed by a slow decrease (Table 2.4). The intercept is the value estimated by the GLMs without

any effect of the predictor variables (pyrene and species interactions) and thus reflects the population

dynamics of D. magna without stress. The initial increase and then the decline of the intercept indicated

that the population was growing until the carrying capacity was reached (Figure 2.1).

27

Table 2.2: Measured pyrene concentrations (μg L-1) for the used pyrene exposure profiles. Measured pyrene concentrations are shown with standard deviations. Nominal concentrations (μg L-1) are shown between brackets when pyrene was added (day 0 and day 8). The letters A, B, C and D refer to the used pyrene treatments.

Experiment 1

Time (d) 0 7 8 10 14 22

Pyrene1A 5.2 ± 0.8 (7.5) 0.2 ± 0.1 10.6 ± 1.4

(15) (3.9 ± 1 0.6 ± 0.2 0.3 ± 0.2

Pyrene1B 13.0 ± 2.2 (20) 0.6 ± 0.3 22.9 ± 4 (40) 12.8 ± 2.1 3.2 ± 0.5 0.3 ± 0.2

Pyrene1C 38.6 ± 11.9 (55) 2.3 ± 0.5 62.8 ±26.6

(110) 44.5 ± 16.9 13.0 ± 2.5 1.5 ± 1.3

Experiment 2

Time (d) 0 2 6 13

Pyrene2A 9 ± 1.4 (20) 4.7 ± 0.5 1.3 ± 0.1 0.3 ± 0.0

Pyrene2B 19.9 ± 2.0 (50) 10.9 ± 0.3 3.2 ± 0.6 0.6 ± 0.0

Pyrene2C 27.4 ± 3.0 (100) 21.4 ± 6.7 6.1 ± 1.6 11 ± 0.5

Pyrene2D 33.9 ± 4.1 (150) 24.5 ± 5.1 8.0 ± 0.9 1.1 ± 0.2

Table 2.3: GLM estimates of pyrene exposure and intraspecific competition for log10-transformed total D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Positive and negative values indicate a higher or lower abundance than the control treatment, respectively. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22

(Intercept) 1.03 1.18 1.64 1.91 2.13 2.03 1.97 1.88

Low pyrene / / / / -0.12 / / 0.10

Medium pyrene / 0.19 / / -0.09 / / /

High pyrene / / / / / -0.07 -0.56 /

Low intraspecific 0.26 0.22 / -0.12 -0.17 -0.09 / -0.07

High intraspecific 0.50 0.39 0.12 -0.11 -0.11 -0.13 / -0.11

Low pyrene X Low intraspecific / / / / 0.17 / / /

Medium pyrene X High intraspecific / / / 0.18 / / / /

High pyrene X High intraspecific / / / / / / 0.30 /

Chapter 2

28

Table 2.4: GLM estimates of pyrene exposure and interspecific competition for log10-transformed total D. magna abundance after backwards model selection for experiment 1 and 2. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Positive and negative values indicate a higher or lower abundance than the control treatment, respectively. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Experiment 1

Time (days) -4 0 2 4 7 10 15 22

(Intercept) 1.06 1.20 1.70 1.90 2.13 2.02 1.97 1.88

Low pyrene / / / / -0.12 / / /

Medium pyrene / 0.17 / / / / / /

High pyrene / / / / / -0.09 -0.63 /

Low interspecific / 0.13 / / -0.15 -0.12 -0.11 /

High interspecific / / / -0.16 -0.24 -0.12 -0.13 /

Low pyrene X High interspecific / / / / 0.17 / / -0.18

Experiment 2

Time (days) -8 -5 -1 2 6 9 13 16

(Intercept) 1.19 1.57 1.95 2.02 2.03 1.97 1.86 1.57

PyreneC1 / / / / / / / /

PyreneC2 / / / / / / -0.38 -0.71

PyreneC3 / / / / / -0.10 -0.34 -0.72

PyreneC4 / / / / -0.18 -0.25 -0.51 -0.88

Competition -0.15 / / / / 0.09 0.25 0.58

Table 2.5: GLM estimates of pyrene exposure and predation for log10-transformed total D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Positive and negative values indicate a higher or lower abundance than the control treatment, respectively. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22

(Intercept) 1.07 1.27 1.69 1.90 2.08 2.01 1.97 1.88

High pyrene / / / / / / -0.56 /

Predation / / -0.32 -0.40 -0.38 -0.28 -0.25 -0.22

High pyrene X Predation / / / / / / 0.37 /

29

Figure 2.1. Total D. magna abundances over time for experiment 1 (A) with four pyrene exposure profiles and for experiment 2 (B) with five pyrene exposure profiles. Data shown are the D. magna population densities with no additional species interactions. Average values with standard deviations (error bars) are depicted. Dashed lines indicate pyrene applications.

(A)

(B)

Chapter 2

30

2.3.3. Effects of pyrene

For the first experiment, the estimated direct effects of pyrene were almost identical between the

different treatments of species interactions (Table 2.3-2.5). For experiment 1, I will therefore only refer

to Table 2.3 here. The first pyrene addition did not significantly affect D. magna population densities

(Figure 2.1 and Table 2.3). However, the highest pyrene exposure did reduce total population densities

7 days after the second pyrene addition (day 15). The description and discussion of the experiment

results will therefore focus on the observed effects after the second pyrene addition. Effects of the

medium and low pyrene exposure profiles on total D. magna abundance were absent or negligible (<

0.13; Table 2.3). The variance of the total population densities explained by pyrene exposure at day 15

was >45% (Appendix A: Table A1-A3). Fourteen days after the second pyrene addition, D. magna

populations were recovering (day 22): no differences in total population densities were observed

between pyrene exposure profiles. However, at that time, the abundances of neonates were higher in the

high pyrene exposure profile than in the control treatment (Appendix A: Table A6 and Figure A2). Also,

the negative effect of high pyrene exposure on the abundances of adults persisted on day 22, although

this effect was smaller than on day 15 (Appendix A: Table A4). Although the total population densities

had recovered, differences in population structure were thus still observed between pyrene treatments

(Figure 2.1 and 2.2). In experiment 2, a similar delay in pyrene effect was observed (Figure 1B).

Significant negative effects were only observed 6 days after the pyrene addition (Table 2.4). Contrary

to experiment 1, significant negative effects were observed at the lower nominal pyrene concentrations.

Also, no recovery was observed at the end of experiment 2 and pyrene effects were visible at all size

classes (Appendix A: Table A13-A15).

2.3.4. Effects of interactions: competition and predation

During the first 9 days of the experiment, the populations with the higher initial population density of

40 individuals (and therefore a higher degree of intraspecific competition) remained more abundant but

this effect decreased with time (Figure 2.3 and Table 2.3). For the populations with an initial population

density of 20 individuals, this effect only persisted during the first 7 days. The variance explained by

intraspecific competition also decreased from 71% to 22% over this period (Appendix A: Table A1). A

high initial density resulted in lower future population densities (starting from day 4), although this

effect was limited (Table 2.3). The population with the lowest initial density (10 Daphnia per test vessel)

reached the highest total D. magna abundance (135 individuals). The initial positive effect of a high

initial density persisted longer for adult D. magna (until day 10; Appendix A: Table A4) than for the

other size classes (day 2 and -4 for juveniles and neonates, respectively; Appendix A: Table A5-A6).

High initial densities resulted in a higher and more constant abundance of adults in the second half of

the experiment compared to low initial densities (Figure 2.2).

31

In experiment 1, B. calyciflorus population densities decreased sharply after one week and B.

calyciflorus completely disappeared by day 10 (Appendix A: Figure A14). Although B. calyciflorus

disappeared, significant but limited differences were observed between population densities of D.

magna of the different interspecific competition treatments starting from day 4 until day 15 (Figure 2.4

and Table 2.4). At the end of the experimental period, differences in population density were no longer

observed between the different degrees of interspecific competition. Abundances of adult D. magna

were never negatively affected by interspecific competition during the whole experiment (Appendix A:

Table A7) while abundances of juvenile and neonate individuals were (Appendix A: Table A8-A9). The

effect of interspecific competition on the abundance of neonates was only significant up to day 10

because almost no neonates were observed in either of the pyrene exposure treatments the following

sample days.

Figure 2.2. Average population structure of D. magna just before and after the second pyrene application (day 8) for different treatments. Data shown are the average abundances of adults (dark grey), juveniles (medium grey) and neonates (light grey) of the specific treatments. “−“ and “+” indicate a significant negative and positive effect, respectively, of that treatment or combination of treatments on total population density, compared to the control treatment (“no pyrene”). High pyrene high pyrene exposure; INTRA = high intraspecific competition treatment; INTER = high interspecific competition treatment; PRED= predation treatment.

Chapter 2

32

Figure 2.3. Total D. magna abundance over time for four pyrene exposure profiles. Data shown are the D. magna population densities for the treatment with no additional species interactions (points), lowintraspecific competition (crosses) and high intraspecific competition (black squares). Average valueswith standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

In experiment 2, B. calyciflorus individuals were added weekly to ensure more continuous competition.

B. calyciflorus population dynamics were similar to experiment 1, with a sharp population decline after

one week (Appendix A: Figure A15). The weekly addition of B. calyciflorus did not result in notable

rotifer densities, although D. magna was significantly affected by pyrene from day 6 onward (Figure

2.4B and Table 2.4). Interspecific competition with rotifers had a negligible negative effect on total D.

magna abundances only on day -8 and even resulted in higher population densities from day 9 onwards

(Figure 4B and Table 2.4).

Of all species interactions studied, predation had the largest negative effect on population densities

(Figure 2.2, 2.5 and Table 2.5). Predation had a continuous negative effect on total D. magna abundance.

The explained variance was always higher than 42%, except on day 15 when most variance was

explained by pyrene exposure (Table 2.5). Because Chaoborus sp. larvae were added 3 days after the

start of the experiment, predation was not significant at day -4. A negative effect of predation was first

observed for adults (at day 0) but the largest effects were observed for the abundances of neonates and

juveniles (Figure 2.2 and Appendix A: Table A10-A12).

33

Figure 2.4. Total D. magna abundance over time for experiment 1 (A) with four pyrene exposure profiles and for experiment 2 (B) with five pyrene exposure profiles. (A) Data shown are the D. magna population densities for the treatment with no additional species interactions (points), low interspecific competition (crosses) and high interspecific competition (black squares); (B) Data shown are the D. magna population densities without interspecific competition (points) and with continuous B. calyciflorus competition (crosses). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the pyrene applications.

(B)

(A)

Chapter 2

34

Figure 2.5. Total D. magna abundance over time for four pyrene exposure profiles. Data shown are the D. magna population densities for the treatment without (points) and with predation (crosses). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

2.3.5. Combined effects of pyrene and species interactions

Significant interactions between pyrene and predation or between pyrene and competition were rare and

most of the times changed inconsistently with increasing pyrene exposure (Tables 2.3-2.5). However,

on day 15, the interaction between high pyrene exposure and predation and between high pyrene

exposure and intraspecific competition positively affected the total D. magna abundance. These positive

interactions indicated that the negative effect of high pyrene exposure was less pronounced when the

population was already exposed to predation or had experienced high intraspecific competition at the

start of the experiment, suggesting antagonism between each of these two types of species interaction

and chemical toxicity. The variance of the total abundance explained by these two interactions on day

15 was 8.4% and 16.8%, respectively (Appendix A: Table A1 and A3).

2.4. Discussion

2.4.1. Pyrene toxicity

Short-term effects of pyrene were limited and the highest effects occurred 7 days after the second pyrene

addition in experiment 1 and 6 days after the pyrene addition in experiment 2 (Table 2.3-2.5). It is

unclear why the first pyrene addition in experiment 1 had no observable effects on population densities.

The highest concentration measured after the first pyrene addition (71 μg/L) was similar to the

EC50,immobilization (68 μg/L), determined in the toxicity test performed on neonate D. magna (<500 μm)

35

from the same clone. However, even neonate D. magna – often considered the most sensitive individuals

(Muyssen and Janssen 2007) - were not affected by the first pyrene addition (Appendix A: Figure A10

and Table A6). The results of the pyrene toxicity test did thus not seem applicable to experiment 1.

It was unclear if the negative effects of pyrene on the abundances of adults after the second pyrene

addition resulted from direct mortality or from a combination of direct mortality and reduced survival

or growth of smaller life classes. Reduced survival and growth of earlier life stages will reduce the

number of juveniles that reach the adult stage (Liess and Foit 2010). The negative effect of pyrene was

largest on abundances of juveniles (Appendix A: Table A4-A6). Adult D. magna were the only size

class still affected by pyrene at the end of the experiment (Appendix A: Table 2.4). Probably, the

negative effect of pyrene on the abundances of adults was thus, at least partly, attributable to effects on

earlier life stages. Neonates were almost absent after the second pyrene addition, even in the control

treatment (Appendix A: Figure A10), which explains the absence of significant pyrene effects for

neonates. A low abundance of small individuals is not uncommon in similar experiments with Daphnia

(Preuss et al. 2009a) and can be attributed to the strong competition which reduces the energy available

for reproduction.

Interestingly, abundances of neonates were significantly higher on day 22 in the high pyrene exposure

profile than in the control treatment (Figure 2.2 and Appendix A: Table A6). As a result, total population

densities were not significantly different between the different pyrene treatments at the end of the

experiment (Table 2.3), leading to the conclusion that total population density was recovered. However,

it should be noted that the final D. magna populations in the high pyrene exposure profile, consisting

mainly of neonates, were probably more susceptible to new chemical stress events than those in the

other pyrene treatments. This illustrates that population structure needs to be accounted for when

assessing the response and recovery of a population to (chemical) stress (Foit et al. 2012).

In the follow-up experiment, measured pyrene concentrations were similar to the measured

concentrations of the first pyrene addition in the original experiment (Table 2.2). Nevertheless, this time

significant effects were observed, even in the lower pyrene concentrations (Table 2.4). It is unclear

whether this is the result of natural variation or of other factors such as the time of pyrene application.

The follow-up experiment was performed six months after the original experiment and it is possible

variations in environmental or biological conditions resulted in the observed differences in pyrene effect.

However, it is likely that the time of application was also an important factor: in the first experiment,

the pyrene was added when the population was growing and competition for resources between

individuals was limited. In the follow-up experiment, pyrene was added after 15 days, when the

population growth phase was already over. Competition for resources between D. magna individuals

was strong at this moment, possibly making the individuals less resistant to pyrene stress.

Chapter 2

36

Pyrene is a substance with low water solubility (0.132 mg L-1). Acetonitrile was used here as a solvent

to ensure that pyrene was dissolved in the water. Solvents are often used with pesticides to ensure a

solution that can easily be applied in the field. Solvents can possibly contribute to the toxicity of a

pesticide and it is therefore important to test the toxicity of the solvent itself. Acetonitrile was not toxic

to D. magna at the concentrations used here.

2.4.2. Species interactions

Of all species interactions studied, predation had the largest effect on total D. magna abundance. Effects

were visible six days after the addition of Chaoborus sp. larvae (starting from day 2) and the highest

effects were observed for the abundances of juveniles and neonates (Figure 2.2 and Appendix A: Table

A10-A12). This indicated a feeding preference: Chaoborus sp. larvae preferred to prey on smaller

juvenile and neonate D. magna than on adult D. magna. Size selective feeding by Chaoborus sp. larvae

has been observed before (Swift 1992). Surprisingly, however, a significant negative effect of predation

was first observed for adults (on day 0) and not for juveniles or neonates. Abundances of juveniles and

neonates were very low (juveniles) or zero (neonates) until day 0, so probably the Chaoborus larvae

were forced to feed on the larger D. magna adults. At later time points, neonates and juveniles were

more abundant and Chaoborus larvae fed on these size classes, leading to a reduced or absent effect of

predation on adults. These data show that feeding preferences depend on the ecological context shaped

by the prey’s population structure.

It is difficult to assess the effects of interspecific competition for the full duration of experiment 1

because B. calyciflorus were reduced to low densities (<10%) after day 7 and completely disappeared

after day 10 (Appendix A: Figure A14). The effects of interspecific competition on total population

densities were therefore limited (Table 2.3). Posterior tests performed with the same conditions showed

that even in the highest pyrene concentration, B. calyciflorus was able to survive for at least 24 days

(Appendix A: Figure A16). The rotifers were thus outcompeted by D. magna, as previously observed in

interaction experiments between B. calyciflorus and D. pulex without chemical stress (Gilbert 1985).

Gilbert et al. (1985) observed limited to no effects of the competition with B. calyciflorus on population

densities of D. pulex, similar to the results of the current chapter. In the follow-up experiment, B.

calyciflorus was added weekly but, even when D. magna was negatively affected by pyrene, notable

population densities of B. calyciflorus were not observed (Appendix A: Figure A15). Clearly, D. magna

was the superior competitor, even when exposed to the pyrene concentrations applied here.

Unexpectedly, weekly addition of rotifers had a positive effect on total D. magna population density in

the second half of experiment 2 (Table 2.4). To avoid the addition of additional algae, rotifers were

taken from the rotifer culture at least 24 hours after feeding. Nevertheless, it seems that algae and

possibly other nutrients were still present in high enough concentrations to positively influence D.

magna population abundances, masking any possible competition effect.

37

Both intraspecific and interspecific competition seemed to result in effects on reproductive output in

experiment 1. Negative effects of different initial densities on the abundances of neonates and juveniles

were observed starting from day 2 while these were absent for adults (Appendix A: Table A4-A6).

Similarly, negative effects of interspecific competition were observed for juveniles and neonates from

day 4 onward while adults were not affected (Appendix A: Table A7-A9). High initial competition thus

mainly affected early life stages at later time points, suggesting competition-induced effects on D.

magna reproduction over direct competition effects. According to the dynamic energy budget (DEB)

theory, the competition with B. calyciflorus or other D. magna individuals could reduce the amount of

energy that could be allocated to reproduction, resulting in less offspring (Kooijman 2010). However, it

is probable that direct competition, through starvation, also contributed to the results. Young D. magna

life stages are more prone to starvation than adults (Preuss et al. 2009a). Under high competition

conditions, less food is available per capita, possibly leading to starvation of smaller individuals and

contributing to the lower proportion of young life stages in the population.

2.4.3. Reduced effect of pyrene when combined with predation and competition

On day 15, when the pyrene effect was largest, predation and intraspecific competition reduced the

negative effect of pyrene on population densities (Figure 2.2). Contrary to the antagonism observed in

the current chapter, species interactions often lead to greater effects of chemical stress. For example, the

combination of predation by Notonecta maculata and exposure to nonylphenol led to loss of resilience

in Daphnia magna populations while individual stressors failed to affect population densities (Gergs et

al. 2013). Synergistic effects of competition and chemical stress have been reported for D. magna (Foit

et al. 2012) and other Daphnia species (Knillmann et al. 2012a). Next to synergistic effects, antagonistic

effects have been reported as well. For example, exposure to predator kairomones led to antagonistic

interactions with carbaryl exposure on reproduction of Daphnia magna (Coors and Meester 2008). This

was attributed to larger-sized and thus more tolerant offspring when predation cues were present. Two

mechanisms are proposed to explain the antagonism observed in the current chapter: differences in

population structure and pyrene-induced alterations in species interactions. First, the structure of the

populations exposed to predation or to high intraspecific competition differed from that of the

populations experiencing low intraspecific competition and populations not exposed to predation. On

day 7, immediately before the second pyrene addition, a large negative effect of intraspecific

competition and predation on the abundances of juveniles and neonates was observed while the

abundances of adults were less affected (Figure 2.2, Appendix A: Table A4-A6;A10-A12). Differences

in sensitivity for different D. magna size classes have been shown before e.g. for four metals (Hoang

and Klaine 2007) or carbaryl (Coors and Meester 2008). Because of the lower proportion of small-sized

and thus more sensitive life stages in populations with predation or high (initial) intraspecific

competition, pyrene effects were smaller. Second, the feeding rate of Chaoborus sp. larvae was possibly

Chapter 2

38

inhibited by the pyrene exposure, leading to reduced predation losses. Indeed, the estimated effect of

predation (Table 2.5) was lower on day 15. The effect of pyrene on the feeding rates of Chaoborus sp.

larvae was not tested in the current chapter but chemicals have been shown to alter feeding behaviour

of fish (Weis et al. 2001) and invertebrates (Maltby and Hills 2008).

Contrary to a similar study with D. magna populations exposed to fenvalerate (Liess and Foit 2010), we

observed no prolonged dominance of smaller-sized organisms after chemical stress in the treatment with

predation: after high pyrene exposure, the proportion of small individuals was higher in the populations

not exposed to predation (Figure 2.2). These contradicting results can be explained by differences in

how predation was applied in the two studies. While Liess and Foit (2010) simulated predation by

removing individuals non-selective on size, Chaoborus sp. larvae preferred to prey on smaller

individuals, leading to lower abundances of neonates in the predation treatments at the end of the

experiment. This highlights the complexity of assessing how ecological interactions alter the response

of a population to chemical stress and the need for ecologically realistic tools (De Laender and Janssen

2013; Gabsi et al. 2014b).

The results presented in the current chapter present an example of how species interactions can lead to

a priori unpredictable effects of chemicals. Predation and intraspecific competition were shown to

interact antagonistically with pyrene when the effect of pyrene was most pronounced. The current

chapter also highlights the need to not only consider the effects of a chemical on population density but

also on population structure when assessing the risk of chemicals for populations and communities.

Although the complexity of the interactions studied in this chapter was limited to interactions between

two species (competitors or predator-prey), this can yield significant insights that are applicable to more

complex food webs. Studies such as this chapter, in combination with approaches such as mechanistic

ecological models (De Laender et al. 2008b; Bontje et al. 2009; Galic et al. 2010), could be used to

integrate species interactions while assessing the long-term ecological risk of a chemical. Using

mechanistic ecological models will increase our understanding on how and when species interactions

influence the effects of a chemical. This will help identify in which situations species interactions are

an important factor for the ecological risk of a chemical and thus when special attention from the

regulators for species interactions is required.

39

3 DEVELOPMENT OF THE DEBKISS IBM

Redrafted from:

Viaene KPJ, De Laender F, Janssen CR, Baveco JM, Van den Brink PJ, Focks A. DEBkiss IBMs as a

novel tool to assess the population-level effects of chemicals on competing species. Submitted

Chapter 3

40

Abstract

Ecological models, in particular individual-based models (IBMs) have been suggested as a way forward

for the ecological risk assessment of chemicals. Most ecotoxicological tests focus on the individual level

and IBMs are ideal tools to translate these test results to the population and community level but need

to be based on a sound theoretical basis. Dynamic energy budget theory based on the keep it simple,

stupid principle (DEBkiss) offers a good compromise between complexity and the amount of data

required to parameterize the model. In this chapter, DEBkiss theory is presented and implemented with

an IBM to describe the life cycle of a generic organism. To account for the effects of chemicals, two

possible toxicity models were implemented: concentration-response curves and toxicokinetic

toxicodynamic models.

DEBkiss IBM development

41

3.1. Introduction

In Chapter 2, I have shown that interactions with other species can be important for population-level

effects in a community context. In the current chapter I will lay the foundations of a food-web model

based on IBMs. To this end, I present an novel and parameter sparse approach to individual-based

modelling based on the DEBkiss theory (Figure 3.1).

Figure 3.1: Overview of the DEBkiss IBM implementation. DEBkiss theory is used to model individual-level processes (see 3.3). An IBM is used to model all individuals in the population and derive population properties (see 3.4). The effects of chemical exposure are modelled using either concentration-response (CR) curves or toxicokinetic-toxicodynamic (TKTD) models (see 3.5).

3.2. Modelling species interactions under chemical stress

The complexity of species interactions makes it difficult to predict population-, community-, and

ecosystem-level effects of chemicals from single-species ecotoxicological tests. Therefore, ecological

models have been used to understand and predict the effects of chemicals on populations (De Laender

et al. 2008b; Galic et al. 2010), communities (De Laender et al. 2014b), and ecosystems (De Laender et

al. 2015). Models can help to test hypotheses regarding the relationship between effects occurring at

lower and higher levels of biological organisation. When based on solid mechanistic understanding, they

can be used to perform experiments in silico. Individual based models (IBMs) seem particularly suited

for use in ERA because population parameters and dynamics emerge from the mechanisms at the

individual level on which most ecotoxicological tests focus (Martin et al. 2013b). IBMs thus allow

exploring how individual-level effects extrapolate to higher levels of organisation (Galic et al. 2014;

Martin et al. 2014). In recent years, IBMs have been used to predict chemical effects on the population

dynamics of a number of typical test species used in ecotoxicology e.g. Daphnia magna (Martin et al.

2013a) and Asellus aquaticus (Galic et al. 2012). However, these IBM applications have focused solely

on the population level and not at higher levels of organisation.

IBMs need a theory describing the underlying individual-level processes. The dynamic energy budget

theory (DEB) was proposed as a suitable underlying theory for the individual level energy balance in

IBMs because this theory provides mechanistic understanding of patterns in growth, reproduction and

Chapter 3

42

mortality of individuals, and enables determination of species-specific parameters (Martin et al. 2013b).

One of the main challenges of standard DEB theory is the number of parameters and the type of

experiments required to accurately parameterize the model (Jager et al. 2013). For many species, these

are not available and the standard DEB model can thus not be parameterized. Also, DEB has some

shortcomings considering its use in a regulatory context: the equations in the standard DEB model can

be difficult to grasp and implementation in software is not straightforward. Therefore, a simplified

version of DEB, DEBkiss, following the paradigm of “keep it simple, stupid” and using less equations

and parameters, has been proposed to allow more transparency, reduce the amount of data required and

increase the user-friendliness (Jager et al. 2013). In DEBkiss, the parameters are more easy to interpret

and they can be more readily related to easily measured endpoints such as the maximum length or the

individual growth rate (Jager et al. 2013). Therefore, DEBkiss IBMs could be a useful tool for species

for which data are sparse.

When IBMs are used to predict the effects of chemicals, the individual-level (physiological) processes

that are sensitive to chemical stress need to be defined. Chemicals can interfere with physiological

processes in many ways. Even in a simplified DEB model, possible physiological modes of action

include effects on assimilation, maintenance and reproduction (Jager and Zimmer 2012). To determine

the actual physiological mode of action of a chemical, ideally full life-cycle studies are used (Jager and

Zimmer 2012). However, when data availability is limited, one is often forced to focus solely on the

effects of a chemical on individual survival. This drastically simplifies the model but also eases its

parameterization because data are more abundant. For example, the General Unified Threshold model

for Survival (GUTS) can be parameterized based on the results of simple acute toxicity tests (Jager et

al. 2011). GUTS is a generalization of a large range of existing toxicokinetic-toxicodynamic (TKTD)

models i.e. models that simulate the time course of processes leading to toxic effects. These models have

the advantage over the often-used concentration-effect equations that they can predict toxic effects even

when exposure is not constant. This allows them to be used in more realistic conditions, when exposure

to a chemical is not constant but changes over time.

3.3. DEBkiss theory

The dynamic energy budget (DEB) theory was originally developed in the ‘80s (Kooijman and Metz

1984) but only became widely applied for making sense of ecotoxicological studies during the past

decade (Kooijman 2010; Jager and Zimmer 2012). In DEB theory, all processes and states of an

individual are expressed as energy (or mass). Next, an energy (or mass) balance for the individual is

created (Kooijman 2010). DEB models describe processes at the level of the individual because,

compared to sub- and supra-individual levels of biological organization, it is relatively easy to make

energy and mass balances at the individual level (Kooijman 2010). DEB models use differential


43

equations to describe how the energy from food is used by the individual for maintenance, growth and

reproduction (Nisbet et al. 2000). A general overview of the mass fluxes considered in the standard DEB

model can be found in Figure 3.2.

The DEBkiss model is based on the standard DEB model but the number of parameters and processes

is reduced by making additional assumptions relating to the life cycle and energy metabolism of the

organism (Jager et al. 2013). The main assumption in DEBkiss is that the energy taken from food is

immediately used to fuel other metabolic processes such as growth, maintenance or reproduction. A

reserve compartment where the energy taken up from food is temporarily stored, is thus not considered

in the DEBkiss theory. The mass fluxes (dry weight per unit of time) in the DEBkiss model are depicted

in Figure 3.2. A concise description of the DEBkiss model used is provided in this chapter. For a more

in depth description of the processes and assumptions in the model, I refer to (Jager et al. 2013).

Abbreviations and symbols for all DEBkiss IBM parameters are provided in Table 3.1.

The typical DEBkiss life cycle consists of three stages: embryo, juvenile and adult. Three compartments

are considered for each individual: the egg buffer , structural body mass and reproduction buffer

. The changes of energy in these compartments over time is described by energy fluxes and are given

by

until (3.1)

(3.2)

(3.3)

The egg buffer is only relevant for embryos and once depleted, the embryo will hatch and become

a juvenile. Once an individual reaches a critical body mass, the individual is considered mature (puberty)

and starts allocating energy to the reproduction buffer.

Chapter 3

44

Fig

ure

3.2:

Gen

eral

sch

eme

of th

e en

ergy

flux

es c

onsi

dere

d in

the

stan

dard

DEB

mod

el (A

) and

in th

e D

EBki

ss m

odel

(B).

(A) I

n th

e st

anda

rd

DEB

mod

el, e

nerg

y fr

om fo

od is

take

n up

(X) a

nd a

ssim

ilate

d in

the

rese

rve

com

part

men

t (A)

. Ene

rgy

from

the

rese

rve

is m

obili

sed

(C) t

o fu

el

som

ati c

mai

nten

ance

(S),

grow

th (G

), m

atur

ity m

aint

enan

ce (J

) and

repr

oduc

tion

(R).

(B) I

n th

e D

EBki

ss m

odel

, ene

rgy

ente

rs th

e in

divi

dual

via

th

e fe

edin

g flu

x (J

X) o

r fro

m th

e eg

g bu

ffer i

f bir

th (b

) has

not

yet

occ

urre

d. T

he a

ssim

ilatio

n flu

x JA

is d

ivid

ed in

a fr

actio

n fo

r gro

wth

(flu

x JV

to

str

uctu

re) a

nd s

omat

ic m

aint

enan

ce (J

M) a

nd a

frac

tion

for

mat

urat

ion

and

repr

oduc

tion

(flux

JR

to th

e re

prod

uctio

n bu

ffer)

, dep

ende

nt o

n w

heth

er p

uber

ty (p

) has

occ

urre

d.


45

Table 3.1: DEBkiss IBM model parameters. Primary parameters refer to parameters that are directly linked to the physiological processes and that do not depend on other parameters. Secondary parameters are determined by one or more primary parameters.

Symbol Parameter Unit

Primary parameters

Maximum specific searching rate L mm-2 d-1

Maximum specific assimilation rate mg dw mm-2 d-1

Volume-specific maintenance costs mg dw mm-3 d-1

Assimilates in a single freshly laid egg mg dw

Structural body mass at puberty mg dw

Yield of structure on assimilates (growth) -

Yield of assimilates on structure (starvation) -

Yield for conversion of reproduction buffer to eggs -

Yield of assimilates on food -

Fraction of assimilates for growth and maintenance -

Conversions

Dry-weight density of structure mg mm-3

Shape correction coefficient -

Fluxes and states

Mass flux for assimilation mg dw d-1

Mass flux for maintenance mg dw d-1

Mass flux for reproduction buffer mg dw d-1

Mass flux for structure mg dw d-1

Mass flux of food mg dw d-1

Mass of egg mg dw

Mass of reproduction buffer mg dw

Mass of structural body mg dw

X Food biomass mg dw L-1

Secondary parameters

f Scaled functional response (0-1) -

Maximum area-specific feeding rate mg dw mm-2 d-1

K Half-saturation food density mg dw L-1

L Volumetric body length mm

Maximum volumetric length mm

Physical body length mm

Von Bertalanffy growth rate constant d-1

ΔR Number of eggs in a clutch number of eggs

Chapter 3

46

Food biomass is taken up by the organism (feeding flux JX) and assimilated by juveniles and adults

(assimilation flux JA) to fuel different metabolic processes. Embryos do not assimilate food but consume

their egg buffer WB until birth. A fraction κ of the assimilation flux JA is used to support maintenance

(JM) and for somatic growth (JV) of the structural biomass WV. A fraction 1 - κ of the assimilation flux

JA is used by juveniles for maturation (not explicitly tracked). After puberty, the 1 - κ fraction of the

assimilation flux describes the mass flux JR towards the reproduction buffer WR. At reproduction events,

the available reproduction buffer is converted to eggs. The number of eggs is determined by the size of

the reproduction buffer. Stochasticity is included in the model in two ways (Martin et al. 2013a). First,

all mortality processes are probabilistic. Second, variation between individuals in DEBkiss parameter

values was included by multiplying the maximum specific assimilation rate with a log-normally

distributed scatter multiplier.

In DEBkiss, food biomass (X) is assumed to be instantly assimilated with a conversion efficiency of

. This conversion efficiency can be used to reflect the quality of the food and a higher efficiency

indicates food of a higher quality. The assimilation flux JA is equal to

with with (3.4)

where f is the scaled functional response that adjusts the assimilation flux to the available food biomass;

is the maximum specific assimilation rate (mg DW mm-2 d-1); L is the volumetric body length (mm);

K is the half-saturation food density (mg DW L-1); is the maximum area-specific feeding rate (mg

DW mm-2 d-1) and the specific searching rate (L mm-2 d-1). Embryos do not consume an external

food source but their egg buffer WB and f is set to 1 for this life stage. Food dynamics can be included

in multiple ways e.g. by daily adding a fixed amount of food or by using a differential equation that

takes into account losses due to feeding.

The change in structural body mass is the result of the amount of energy that is assimilated minus the

amount of energy needed for maintenance. The flux in structural body mass is thus equal to

where (3.5)

where JV is the mass flux to structure (mg DW d-1); is the yield of structure on assimilates; κ is the

fraction of assimilates for growth and maintenance; JA is the mass flux for assimilation (mg DW d-1); JM

the mass flux for maintenance (mg DW d-1); the volume-specific maintenance costs (mg DW mm-3

d-1) and L the volumetric length (mm).

The growth of an organism can be simplified to a von Bertalanffy equation (Jager et al. 2013) equal to

with and (3.6)


47

where L is the volumetric length (mm); rB the von Bertalanffy growth constant (d-1); is the yield of

structure on assimilates; dV the dry-weight density of structure; the volume-specific maintenance

costs (mg DW mm-3 d-1); Lm the maximum volumetric length an individual can obtain (mm); κ is the

fraction of assimilates for growth and maintenance; is the maximum specific assimilation rate (mg

DW mm-2 d-1) and the volume-specific maintenance costs (mg DW mm-3 d-1).

The (1 – κ) fraction of the assimilation flux JA is used for maturation when the individual is juvenile

( ) or for the reproduction buffer when the individual is mature. The dynamics of the mass in

the reproduction buffer JR (mg DW d-1) are described by

(3.7)

where κ is the fraction of assimilates for growth and maintenance and JA the mass flux for assimilation.

At reproduction events, the available reproduction buffer is transformed into a number of eggs ( )

equal to

(3.8)

where is the yield for conversion of reproduction buffer to eggs; floor indicates that the lowest

whole number made possible by the division is selected; WR the mass of the reproduction buffer (mg

DW) and WB0 the egg mass (mg DW).

A special set of rules was used in the case of starvation (Jager et al. 2013). Initially, the energy shortage

is taken from the reproduction buffer (if any). In a second phase of starvation, the energy needed is

translated from the structural biomass and the organism shrinks. If an organism shrinks to a critical

fraction of its maximal attained weight, it has a high chance (0.35 d-1) of dying.

first stage: if but or

(3.9)

(3.10)

second stage: if and and

(3.11)

(3.12)

where κ is the fraction of assimilates for growth and maintenance; JA the mass flux for assimilation (mg

DW mm-2 d-1); JM the mass flux for maintenance (mg DW mm-2 d-1); WR the mass of the reproduction

buffer (mg DW); JV the mass flux for structure (mg DW mm-2 d-1) and the yield of assimilates on

structure (starvation).

Chapter 3

48

3.4. The DEBkiss individual based model

A general feature of an IBM is that at every time step, all individuals obey a series of sequential rules

(Figure 3.3). A specific feature of the DEBkiss IBM developed here is that these rules are based on

DEBkiss theory (see 3.2). First, the individual takes up energy from food. Next, mortality of the

individual is evaluated: does the individual die from stochastic processes, starvation, toxicity or aging?

If the individual survives, the growth of the individual is calculated. If the individual is not yet adult, no

more rules are considered this time step and the individual continues to the next time step. If the

individual is adult, the next rule is to evaluate whether reproduction will occur. If not, the reproduction

buffer is updated and the individual continues to the next time step. If reproduction occurs, the number

of offspring is determined and afterwards the reproduction buffer is updated. New-born individuals

undergo the same rules in the next time step.

Figure 3.3: Rules that an individual in the IBM goes through during a time step. Rules that change their outcome depending on the state of the individual are marked in grey.

3.5. Toxicity implementation

Toxicity can be included in multiple ways in DEB model simulations. Often, toxic effects in DEB

models are described by impairment of one, or a combination of, different physiological modes of action,

where effects on survival, assimilation, growth, maintenance and reproduction are the most common

(Kooijman 2010; Jager and Zimmer 2012). In this modelling framework, chemicals can thus have a

wide range of effects on organisms and their physiology, and it can be very difficult to discern which

modes of action are relevant for a specific chemical. Moreover, it is unclear how sub-lethal effects


49

translate to the population level. Consistent with the philosophy of DEBkiss, a simplified approach was

implemented where only effects on survival were considered. Two approaches to predict the effects of

a chemical on survival were evaluated: concentration-response (CR) relationships and toxicokinetic-

toxicodynamic (TKTD) models.

3.5.1. Concentration-response relationships

CR relationships give a straightforward relationship between the concentration of the chemical in the

water and the survival probability of an individual, based on standard toxicity test results. The survival

probability for a given time period (e.g. one day) can be predicted from the water concentration by

(3.13)

which only includes the concentration of the chemical in the water ( ) and two chemical-specific

parameters: the slope parameter and the of the chemical.

3.5.2. Toxicokinetic-toxicodynamic models

The TKTD model used is based on the general unified threshold model for survival (GUTS) which

offers a general framework for the wide variety of TKTD models available (Jager et al. 2011). In short,

toxicokinetic-toxicodynamic models predict the internal concentration over time from a given external

concentration (toxicokinetics) and link this internal concentration to a toxic effect at the level of the

individual over time (toxicodynamics) (Jager et al. 2011). The implemented TKTD model is based on

the reduced GUTS-SD model, assuming stochastic death and using a scaled internal concentration

(Figure 3.4). The GUTS-IT model, assuming the individual tolerance concept, was also implemented

but was sub-optimal for the species and chemical used here when applied to standard toxicity

experiments.

Figure 3.4: The implemented TKTD model. The dominant rate constant (kD1) describes the rate at which a chemical is taken up by the individual (C*int) from the environment (Cext). The killing rate constant (kk) and the internal threshold for effects (z) are used to link the scaled internal concentration (C*int) to a survival probability over time.

The TKTD model is described by three parameters: the dominant rate constant kD, the killing rate

constant kK and the internal threshold for effects z. To calculate the internal concentration of a chemical

(toxicokinetics), a first-order one-compartment model can often be used:

Chapter 3

50

(3.14)

where Cint is the internal concentration (μg kg-1); t is the time (d), kI is the uptake rate constant (L kg-1 d-

1); Cext the external concentration (μg -1) and kE the elimination rate constant (d-1). However, in

ecotoxicological studies, the internal concentration is mostly not measured. The GUTS model can still

be parameterized then by rescaling equation (3.14) and dividing both sides by kI/kE. This way, the

rescaled internal concentration C*int is directly proportional to the actual internal concentration and has

the dimensions of the external concentration:

(3.15)

where is the rescaled internal concentration (μg L-1); t is the time (d); kD is the dominant rate

constant and Cext the external concentration (μg -1). The elimination rate constant can be estimated from

the time-course of survival data even if internal concentrations are unknown. However, it is not

guaranteed that this estimated elimination rate constant represents only whole-body elimination. When

the elimination rate is estimated from only external concentration and survival data over time, the

slowest compensatory process, which can be either toxicokinetic (i.e. elimination from the body) or

toxicodynamic (i.e. recovery from damage), will dominate the dynamics of toxicity. Therefore, we refer

to the dominant rate constant kD instead of the elimination rate constant kE for the rescaled internal

concentration (Equation 3.15). To translate this rescaled internal concentration to effects on survival

over time, the cumulative hazard was used:

with (3.16)

where S(t) is the survival probability of an individual at time t; t is the time (d); HZ the cumulative hazard

of an individual at time t; kK the killing rate (L μg-1 d-1); selects the maximum of 0 and ;

is the rescaled internal concentration (μg L-1); z the internal threshold for effects (μg L-1) and hB the

background mortality rate (d-1).

Both toxicity approaches provide the probability, based on the chemical exposure, that an individual

will survive one more day. This survival probability is then evaluated for each individual in a population

to determine if it survives until the next day. The main difference between the two toxicity sub-models

is that CR relationships only consider the current chemical concentration and not earlier exposure while

chemical effects can still occur after exposure when using TKTD-SD.

3.6. Software

The DEBkiss IBM framework was implemented in NetLogo (Wilensky 1999), which was specifically

designed for IBMs and has been used before to implement DEB IBMs (Martin et al. 2012).

51

4 APPLICATION OF THE DEBKISS IBM FRAMEWORK TO

ASSESS THE EFFECTS OF COMPETITION AND CHEMICAL

STRESS ON THE POPULATION DYNAMICS OF

BRACHIONUS CALYCIFLORUS AND DAPHNIA MAGNA

Redrafted from:

Viaene KPJ, De Laender F, Janssen CR, Baveco JM, Van den Brink PJ, Focks A. DEBkiss IBMs as a

novel tool to assess the population-level effects of chemicals on competing species. In preparation.

Chapter 4

52

Abstract

To increase the ecological relevance of chemical risk assessment, the potential influence of species

interactions such as competition need to be accounted for. Ecological models have been identified as a

possible approach to consider species interactions, however, high data requirements often impair the

use of modelling approaches. Individual-based models based on a simplified dynamic energy budget

theory (DEBkiss IBMs) provide transparent and readily applicable models. In the current chapter, I

tested the capacity of an integrated model composed of two DEBkiss IBMs to predict the effects of pyrene

on two competing consumers, the cladoceran Daphnia magna and the rotifer Brachionus calyciflorus.

Both IBMs were calibrated using only data available in the literature and their performance was

evaluated using data from competition experiments. To predict the effects of pyrene, two possible toxicity

model implementations – concentration-response curves and toxicokinetic-toxicodynamic models – in

the IBMs were compared. Population dynamics of both species in isolation were reasonably well

predicted, although the population structure for D. magna was less well predicted. Agreeing with the

results of the experiments, D. magna outcompeted B. calyciflorus in the integrated model with both

species. The toxicokinetic-toxicodynamic model captured short-term effects of pyrene better than the

concentration-response model but overpredicted the chronic effects. Simulated rotifer population

dynamics exposed to both pyrene and competition were higher than those observed because of the

overprediction of pyrene effects on daphnid populations. Despite its limitations, the tested DEBkiss IBM

approach can be a useful tool for assessing the risk to species with short life cycles in case of low data

availability.

DEBkiss IBM application to experimental results

53

4.1. Introduction

Ecological models and especially individual-based models (IBMs) have received increasing attention in

ecotoxicology during the past decade. IBMs have been developed for several species such as Asellus

aquaticus (Galic et al. 2012), Daphnia magna (Preuss et al. 2009a) and Chaoborus (Dohmen et al.

2015). However, these models are still met with much scepticism by industry and regulators that are

unfamiliar with using models for effect assessment. In order to improve their use and acceptance, models

need to be validated (Grimm and Martin 2013; Grimm and Thorbek 2014). One common way to validate

models is to compare model output with patterns in available data. Patterns in this context are defined

as characteristics of the observations that are unlikely to be the result of random processes (Grimm and

Martin 2013). For complex systems, it is critical to evaluate multiple patterns because single patterns

can often be reproduced in multiple ways and it is then unknown whether the underlying mechanisms

are correctly captured. Key characteristics of complex ecological systems are interactions between

individuals of different species. Ecological models in general and IBMs more specifically that can

accurately predict how these interactions affect the population dynamics of the species involved,

especially in the context of chemical stress, could become essential tools in future ecological risk

assessment approaches.

The main objective of this chapter was therefore to test whether the DEBkiss IBM framework described

in Chapter 3 is suitable to simulate and understand patterns observed for two interacting populations

under chemical stress. To reach this objective, DEBkiss IBMs were implemented and parameterized for

two grazers (the water flea Daphnia magna and the rotifer Brachionus calyciflorus) and a framework

was developed to allow multiple DEBkiss IBMs to compete for a shared resource (the algae

Scenedesmus obliquus). Simulations were performed for the control treatment, for exposure to pyrene,

for interspecific competition, and for a combination of pyrene exposure and interspecific competition.

To evaluate the performance of the model for predicting the combination of ecological (competition)

and chemical stress, simulation results were compared to the patterns observed in the experiments

described in Chapter 2.

4.2. Materials and Methods

4.2.1. Model parameterization

Available literature data were used to determine a range of possible DEBkiss IBM parameter values

(Table 4.1). The available literature for Daphnia magna and Brachionus calyciflorus was queried for

DEB-related parameters such as maximum body lengths, growth rates, ingestion rates, egg weights and

length-weight relationships. Parameter values from other DEB models were avoided because these have

been fitted for more complex DEB models which differ from DEBkiss (e.g. by including a reserve

Chapter 4

54

compartment) and were therefore incomparable to the DEBkiss parameters. If not found, DEBkiss

parameters were calculated from the available information. For example, if the von Bertalanffy growth

constant ( ) was known and standard values of 0.1 mg DW mm-3 and 0.8 were used for the dry-weight

density of structure ( ) and the yield of structure on assimilates ( ), respectively, the volume-specific

maintenance costs were calculated using equation (3.6). This approach assumed that (1) the available

literature data were applicable to the individuals used in the current chapter, which, given the large

variation within a species, is not always certain and (2) the assumptions made in the original DEBkiss

paper also applied here. Both the concentration-response curves and the TKTD model were

parameterized using a standard 48-hour toxicity test performed with D. magna (See Chapter 2).

4.2.2. Species-specific adjustments to the DEBkiss-IBM framework

The DEBkiss IBM framework offers a very general description of an organism’s life cycle. This is very

useful when specifics about the life cycle and physiology are not available. In the current chapter, two

species are considered that have been widely used and studied in ecotoxicology. Species-specific details

are thus known and species specific traits can be accounted for in the DEBkiss IBM.

Daphnia magna

Because of their carapace, D. magna individuals cannot physically shrink when starving (Martin et al.

2013a). So, although their weight decreases, their length stays constant. This has implications for the

assimilation of food because the amount of food daphnids obtain from the water depends on the length

of the appendages. Therefore, instead of using the actual length of an individual, the maximum obtained

length was used to calculate the assimilation flux (Equation 3-4).

Under starvation, daphnids can convert body mass to energy until a critical weight is attained and

mortality occurs (Rinke and Vijverberg 2005). To simulate this in the current chapter, individuals that

shrink to 40% of their previously attained maximum weight have a high probability of dying (per capita

death rate of 0.35 d-1 (Martin et al. 2013a)). Similar starvation rules were found insufficient in an earlier

DEB IBM implementation for D. magna (Martin et al. 2013a). Indeed, preliminary simulations with

these standard starvation rules showed shortcomings similar to those reported in Martin et al. 2013a

study: densities of individuals in the medium size class were overpredicted and no increase of

individuals in the largest size class was predicted when food becomes limiting.


55

Bra

chio

nus c

alyc

iflor

us

[0.0

2 –

0.12

] * 1

0-3 m

g dw

(D

umon

t et a

l. 19

75; S

aund

ers I

II a

nd

Lew

isJr

1988

)[0

.15

– 0.

21] *

10-3

mg

dw a

[0.0

88 –

0.1

95] m

g dw

mm

-2 d

-1 a

[0.3

37 –

0.7

47] L

mm

-2 d

-1 a

0.76

47 m

g dw

mm

-3 d

-1 a

2.03

93 d

-1 a

[0.0

92 –

0.2

04] m

m a

0.32

6391

4 dw

L-1

(Han

sen

and

Bjo

rnse

n 19

97)

[0.6

-0.9

5] b

0.8 b

0.95

b

[0.6

-0.9

5] b

[0.4

– 0

.9] b

0.1

mg

dw m

m-3

a

0.50

97 a

0.5

d (H

alba

ch 1

970b

) 0.

8 a

0.01

d-1

a

10 d

(H

alba

ch 1

970b

)

Dap

hnia

mag

na

[7-1

0] *

10-3

mg

dw

(Dum

ont e

t al.

1975

; Till

man

n an

d La

mpe

rt 19

84;

Gla

zier

1992

)[0

.027

– 0

.031

5] m

g dw

a

[0.0

6 –

0.12

1] m

g dw

mm

-2 d

-1 a

[0.0

265

– 0.

6526

] L m

m-2

d-1

a

[0.0

4125

– 0

.051

] mg

dw m

m-3

d-1

a

0.13

7 d-1

(Pie

ters

et a

l. 20

06; J

ager

and

Zim

mer

201

2)

1.9

mm

a

[0.1

22 –

2.2

38] m

g dw

L-1

(Mul

der a

nd H

endr

iks 2

014)

0.8 b

0.8 b

0.95

b

0.8 b

[0.5

– 0

.9] b

0.1

mg

dw m

m-3

a

0.38

a

2.8

d a

0.4

a

0.02

d-1

a

96 d

(M

acA

rthur

and

Bai

llie

1929

)

Para

met

er

Ass

imila

tes i

n a

sing

le fr

eshl

y la

id e

gg

Stru

ctur

al b

ody

mas

s at p

uber

ty

Max

imum

spec

ific

assi

mila

tion

rate

Spec

ific

sear

chin

g ra

te

Vol

ume-

spec

ific

mai

nten

ance

cos

ts

Von

Ber

tala

nffy

gro

wth

rate

con

stan

t

Max

imum

vol

umet

ric le

ngth

Hal

f-sa

tura

tion

food

den

sity

Yie

ld o

f stru

ctur

e on

ass

imila

tes (

grow

th)

Yie

ld o

f ass

imila

tes o

n st

ruct

ure

(sta

rvat

ion)

Yie

ld fo

r con

vers

ion

of re

prod

uctio

n bu

ffer

to e

ggs

Yie

ld o

f ass

imila

tes o

n fo

od

Frac

tion

of a

ssim

ilate

s for

gro

wth

and

mai

nten

ance

Dry

-wei

ght d

ensi

ty o

f stru

ctur

e

Shap

e co

rrec

tion

coef

ficie

nt

Tim

e be

twee

n re

prod

uctio

n ev

ents

Bod

y m

ass f

ract

ion

at w

hich

star

vatio

n oc

curs

Stoc

hast

ic m

orta

lity

rate

Max

imum

age

of a

n in

divi

dual

Sym

bol

K

- - - -

Tabl

e 4.

1: P

aram

eter

val

ue r

ange

s fo

r D

. mag

na a

nd B

. cal

ycifl

orus

. a C

alcu

late

d us

ing

equa

tions

(3.4

) and

(3.6

). b

Sugg

este

d va

lues

Chapter 4

56

Therefore, I evaluated if model fit could be improved by additional starvation rules. In Martin et al.

(Martin et al. 2013a), this problem was mitigated by including an additional mortality term based on the

amount of energy in the reserve compartment. In DEBkiss, a reserve compartment is missing (Jager et

al. 2013) and therefore a mortality rate based on the reserve density cannot be included. However, this

additional mortality can be viewed as a time-weighted average of the feeding history (Martin et al.

2013a), summarized in the functional response f in DEBkiss. Therefore, a time-weighted average of the

functional response values of the last five days is calculated (fH) and combined with a mortality

coefficient m to provide a starvation-dependent additional mortality probability:

Probability(mortality) = m (1 – fH) (4.1),

Earlier work has shown that only considering this starvation mortality for juveniles led to the best

simulations (Martin et al. 2013a). In the current implementation, starvation mortality was therefore only

implemented for juveniles and values between 0 and 0.2 were evaluated for the mortality coefficient m.

Brachionus calyciflorus

B. calyciflorus is known to have a post-reproductive period of 1-2 days (at 20°C) where individuals still

feed but don’t reproduce (Jensen and Verschoor 2004). Given that the typical life duration of B.

calyciflorus is 10-11 days at 20°C (Jensen and Verschoor 2004), this is quite significant. This post-

reproductive period was included by limiting the reproduction to individuals younger than 9 days. Also,

a maximum age of 11 days was included for rotifers. Post-reproduction individuals were not eliminated

from the simulated populations because they still feed and thus influence other B. calyciflorus and D.

magna individuals.

4.2.3. Coupling

Species that are in competition need to be able to interact with one another. Instead of implementing

both species in one NetLogo instance, we opted to write a central interface in Java that calls separate

NetLogo instances. This allows the whole modelling framework to be more flexible, making it easier to

e.g. adjust the food input, use species-specific DEBkiss IBMs and extending the framework to more

than two species. This implementation also made it easier to the expand the food web further, as

described in chapter 5-6.

4.2.4. Comparison with data from experiments

DEBkiss IBM simulations were compared to the experimental results described in Chapter 2. In these

experiments, isolated populations of B. calyciflorus and D. magna and the two species in competition

were exposed to a pyrene pulse. Using the population dynamics of the control populations (without

competition and pyrene stress), parameter values were optimized. For D. magna, control population


57

dynamics were available from two experiments (see Chapter 2) and both were used for parameter

optimization. To account for variation in the parameters, 100 simulations were performed with random

sampling from the range of possible DEBkiss parameter values given in Table 4.1. 100 simulations were

chosen because this was deemed a good compromise between computation time and accounting for the

variation on the DEBkiss parameters. Optimization was done by likelihood optimization (Jager and

Zimmer 2012) with the likelihood l of the parameter set θ given the data Y calculated as

where N is the total number of data points and SSQ the sum of squared residuals. The SSQ is given by

(4.3)

where i is the time point of the observation (from i to k), the estimated value for observation i using

parameter set θ and the observed value for observation i.

Population dynamics of the control treatments were used to select the 10 best (out of 100) parameter

combinations for each species based on the calculated likelihoods (Table 4.2). Because the size structure

of the population was important for the response to stress (Chapter 2), likelihoods for D. magna were

calculated for both the total density and the densities of the size classes. The likelihoods of the total

density and of the three size classes were then multiplied to derive one value to be optimized. By doing

so, the parameter combinations were selected that best described both the total population density and

the population structure. Initial exploration of the likelihoods showed that the maximum area-specific

assimilation rate ( ) was the dominant factor determining the likelihood, with the lowest likelihood

for high values (Appendix B Figure B1). Therefore, the range for was limited in the simulations to

values between 0.06 mg mm-2 d-1 and 0.121 mg mm-2 d-1. Extinction of B. calyciflorus in the control

treatment was observed by the end of the experiment (31 days), most likely due to abiotic causes e.g.

decreased water quality. Therefore, only the population densities of the control treatment of the first 22

days were used to fit the model.

Table 4.2: Steps, parameters and experiments used for the simulations in this Chapter.

Step Parameters used Related Chapter 2 experiment 1 100 random combinations of literature

parameter values (Table 4.1) Experiment 1: Control Experiment 2: Control

2 10 optimal parameter combinations from step 1 Experiment 2: Interspecific competition 3 10 optimal parameter combinations from step 1 Experiment 2: Pyrene 4 10 optimal parameter combinations from step 1 Experiment 2: Interspecific competition +

Pyrene

Chapter 4

58

The 10 optimal parameter combinations for both species were then used to predict the dynamics for

exposure to competition, exposure to pyrene and exposure to both competition and pyrene (Table 4.2).

When simulating two competing species, each of the 10 optimal parameter combination of D. magna

were combined with each of the 10 optimal parameter combination of B. calyciflorus, resulting in a total

of 100 simulations. To evaluate the effects of pyrene, both the concentration-response (CR) and the

toxicokinetic-toxicodynamic (TKTD) models were tested. Simulations were only compared to the data

of experiment 2 because it was unclear how exposure to two pyrene pulses, as was the case in experiment

1, affected the daphnids and if the toxicity models could capture this. In experiment 2, only one pyrene

addition was performed and the applicability of both toxicity models should be more clear. 100

simulations were run for the pyrene exposure to account for stochasticity. Simulations for exposure to

both competition and pyrene were also simulated 100 times. For all these treatments, the calculated

likelihoods were used again to select the 10 simulations (out of 100) closest to the observed densities.

4.3. Results

4.3.1. Daphnia magna

Figures 4.1-4.3 in the current chapter show the optimal fit of the DEBkiss IBMs i.e. the 10 best

simulations of the 100 simulations performed per treatment, based on the likelihood. Simulations of

isolated D. magna populations without chemical stress showed a population growth phase followed by

a decline as the carrying capacity of the system was reached (Figure 4.1). Despite differences in the

observed maximum population densities of the two experiments used for parameter optimization (150

individuals versus 110 individuals), the population trends in both experiments were similar (Figure 4.1).

The maximum density of the first experiment was predicted by some of the 100 performed simulations

but since the optimization was performed for all counted abundances, the best simulations shown in

Figure 4.1 are more in line with the maximum density counted in the second experiment. The maximum

population density in the simulations was lower and occurred earlier than in the experimental

observations. In the second half of the experiments, the simulated population density was therefore lower

than the observed density. The population densities at the end of the experiment were more comparable

between simulations and experiments (Figure 4.1).

The largest discrepancy between the control treatment of experiments and simulations was observed for

the abundance of large individuals at the end of the experiment (Appendix B Figure B2). While a steady

increase of large individuals was observed until the end of the experiment, a constant, low abundance

of large individuals was predicted. Related to this, the abundance of intermediate individuals strongly

decreased in the experiments while this was not observed in the simulations. A similar trend was

observed in an earlier study and was alleviated by including additional starvation mortality related to

the reserve density of an individual (Martin et al. 2013a). Because DEBkiss IBMs lack a reserve


59

compartment, additional starvation mortality based on the history of the functional response parameter

f was used as an alternative. Adding this functional response based starvation mortality did, however,

not decrease the deviation for large individuals nor increase the fit for total abundances (Appendix B

Figure B2-B3). Therefore, this additional starvation mortality mechanism was not used in other

simulations shown in the current chapter (Figure 4.2-4.3). Similar to the experiments, the abundance of

small individuals, indicating reproduction, was highest in the first half of the simulation period.

However, while reproduction was not observed after day 10 in the experiments, low abundances (<20

individuals) of small individuals were still observed in the simulations.

Figure 4.1: Observed and predicted population dynamics for D. magna using a DEBkiss IBM. D. magna populations were exposed to (A) pyrene; (B) the competitor B. calyciflorus and (C) the competitor B. calyciflorus and pyrene. The grey area indicates the 10 best simulations out of 100 for the control treatment (no pyrene exposure or competition). Green shaded areas indicate the 10 best simulations out of 100 for populations under stress. Black bullets and squares show the observed dynamics for the control treatment in the two experiments. Green squares show the observed dynamics for populations under stress (competition and/or pyrene exposure). The red dashed line indicate the time of pyrene addition. Asterisks indicate significant differences (p < 0.05) between control and treatment in the experiment.

Chapter 4

60

Two toxicity models were evaluated, concentration-response curves (CR) and toxicokinetic-

toxicodynamic models with stochastic death (TKTD-SD; Figure 4.2). Effects of pyrene were only

considered for D. magna as no effects were observed for B. calyciflorus, neither in the acute toxicity

test (Chapter 2) nor in the population experiment (Figure 4.3). Using CR curves, pyrene effects occurred

immediately while a delay of a week was observed in the experiments (Figure 4.1). Also, D. magna

populations recovered after two weeks in the simulations while this was not observed in the experiment.

Using TKTD-SD, effects were more delayed (1-2 days) but still too fast. Also, the effects of pyrene

were larger, potentially even leading to extinction of the daphnid population, and persisted longer than

with the CR toxicity model (Figure 4.2). Both considered toxicity models were thus unable to fully

capture the effects of pyrene on D. magna population dynamics. Although the predicted effects of pyrene

were too high, the TKTD-SD model was preferred over the CR model because the slight delay in pyrene

effects and the absence of population recovery matched the observations most closely.

Figure 4.2: Comparison of the DEBkiss simulations for D. magna with concentration-response curve (A) or TKTD-SD (B) as toxic effect sub-model. The grey area indicates the 10 best simulations out of100 for the control treatment (no pyrene exposure or competition). Green shaded areas indicate the 10best simulations out of 100 for populations under stress. Black bullets and squares show the observeddynamics for the control treatment in the two experiments. Green squares show the observed dynamicsfor populations under stress (competition and/or pyrene exposure). The red dashed line indicates thetime of pyrene addition. Asterisks indicate significant differences (p < 0.05) between control andtreatment in the experiment.


61

4.3.2. Brachionus calyciflorus

Simulations for the isolated B. calyciflorus populations reflected the pattern of population increase and

decrease observed in the experiment (Figure 4.3). For B. calyciflorus, the simulated maximum total

population density was comparable to that observed in the experiments. Because pyrene effects were

not predicted for individuals at the measured pyrene concentrations, exposure to pyrene did not alter the

B. calyciflorus population dynamics.

Figure 4.3: Simulated population dynamics of B. calyciflorus using a DEBkiss IBM. B. calyciflorus populations were exposed to (A) pyrene; (B) competition with D. magna and (C) both competition with D. magna and pyrene exposure. The grey area indicates the 10 best simulations out of 100 for the control treatment (no pyrene exposure or competition). Green shaded areas indicate the 10 best simulations out of 100 for populations under stress. Black squares show the observed dynamics for the control treatment in the two experiments. Green squares show the observed dynamics for populations under stress (competition and/or pyrene exposure). The red dashed lines indicates the time of pyrene addition. Asterisks indicate significant differences (p < 0.05) between control and treatment in the experiment.

4.3.3. Competition

As observed in the experiments, daphnids were also the superior competitors in the simulations (Figure

4.1 and Figure 4.3). Simulated daphnid population dynamics were largely unaffected by competition

with rotifers, which corresponds to the experimental results. The only effects of competition on daphnid

populations were a 23% lower maximum population density and an earlier cessation of population

growth (five days earlier) in the competition treatment simulations than in the simulations of the control

treatment. However, differences between isolated populations and populations in competition with

rotifers were not observed in the experiment. Simulated population dynamics of B. calyciflorus

competing with D. magna corresponded well with experimental results (Figure 4.3): a population

Chapter 4

62

increase when food was not yet limiting – but lower than for isolated B. calyciflorus populations – and

a very low population abundance or extinction when food was scarce, although the population persisted

longer than observed.

4.3.4. Competition and chemical stress

When D. magna was exposed to both pyrene and competition, the effect of pyrene on D. magna was

unaltered: daphnid populations decreased, potentially leading to extinction (Figure 4.1). This was similar

to that observed in the experiments where the effect of pyrene showed no interaction with the effect of

interspecific competition (see Chapter 2). Contrary to the experiments, rotifers increased in abundance

after pyrene exposure (Figure 4.3).

4.4. Discussion

4.4.1. Predictive capability of DEBkiss IBM for Daphnia magna and Brachionus

calyciflorus

In general, the effects of food shortage on D. magna populations appeared sooner in the DEBkiss IBM

than in reality. Several deviations between the experiments and the simulations could be ascribed to the

absence of a reserve compartment in the D. magna DEBkiss IBM: compared to the experiments, the

simulations showed lower maximum population densities, earlier population decline, lower abundance

of large individuals and higher effects of competition with B. calyciflorus. The implementation of an

additional starvation mechanism similar to another D. magna DEB IBM implementation (Martin et al.

2013a) proved unsuccessful. The absence of a reserve compartment thus limits the application of the

DEBkiss IBM for conditions when the population is starving. The starvation recovery rules were also

lacking: while reproduction was not observed after day 10 in the experiments, low abundances (<20

individuals) of small individuals were still observed in the simulations. In the DEBkiss implementation,

starving individuals shrink to meet their energy needs. When food is available again, they immediately

spend a fixed part of the assimilated energy on reproduction. However, it seems unlikely that the

proportion of assimilated energy spent on reproduction is identical for well-fed individuals and

individuals recovering from starvation. Implementing starvation recovery rules that favour recovery of

weight over growth and reproduction could improve the fit of the simulations. However, many different

starvation recovery strategies have been proposed for D. magna and a general consensus is lacking

(Vanoverbeke 2008).

The instantaneous effects of pyrene on D. magna populations when modelling toxicity with CR curves

is not surprising given that only the current chemical concentration is used to predict pyrene effects.

There is thus no delay between exposure and effects possible, although a delay of about a week was


63

observed for pyrene. Similarly, effects of pyrene disappear quickly with the CR approach, explaining

the predicted recovery. The TKTD approach can account for a delay in effects and recovery but effects

were still too fast for the experiments used. A possible cause for the deviation between the effects

predicted by the toxicity models and the observed effects is that the parameterization of these models

was done using the result from an acute toxicity test performed with D. magna juveniles. Chronic

toxicity can differ from acute toxicity e.g. because of alternative modes of action or toxicity of the

biotransformation products (Dom et al. 2012). Also, the endpoint used in the D. magna toxicity test was

immobilization and it can take a few days for immobilized individuals to die. Moreover, the presence of

different size classes can influence the susceptibility of a population to toxic stress: larger D. magna

individuals have been shown to be more resistant to e.g. carbaryl (Coors and Meester 2008), possibly

explaining the observed delay in effect of pyrene at concentrations close to the acute EC50. Alternatively,

due to its assumed narcotic mode of action (Di Toro et al. 2000), it is possible that pyrene causes sub-

lethal effects in daphnids. Including such sub-lethal effects might explain the delayed effect of pyrene

on total population density. More complex toxicity models which e.g. consider effects on maintenance

costs or other modes of action were not tested because (1) the aim of the chapter was to test the efficacy

of a simple (toxicity) model and (2) accurate information about the mode of action of pyrene or different

sensitivity of life stages to pyrene is lacking.

The observed population decline of rotifers at the end of the experiment was most likely due to

unsuitable environmental conditions e.g. decreased water quality, waste accumulation and/or changed

physico-chemical parameters of the water, which are not considered in the IBMs. For example, the pH

decreased from 7.7 ± 0.1 at the start to 6.5 ± 0.3 at the end of the second experiment and the oxidation-

reduction potential increased over the same period from -61 ± 5 mV to 9 ± 16 mV. Other water quality

parameter and waste products were not measured in the experiments.

Competitive exclusion has been observed frequently for rotifers competing with daphnids (MacIsaac

and Gilbert 1989). We were able to simulate this exclusion and its effects on both species, although the

disappearance of rotifers was slower in the simulations than in the experiments. Possibly, other

mechanisms not considered here accelerated the extinction of rotifers e.g. rotifers can be damaged by

mechanical interference – rejection after accidental ingestion – with feeding daphnids (Gilbert 1988). It

is also plausible that such low abundances of B. calyciflorus were simply not detected in the experiments.

Because of the transparency and low complexity of the DEBkiss IBMs, it would be easy to implement

additional competition mechanisms for species when this is needed.

The predicted increase in rotifers after pyrene addition did not correspond with the experiments but is

logical when looking at the other simulations. This difference was caused by the rotifers still being

present in the simulations at the time of pyrene addition contrary to the experiments. Moreover, the

effect of pyrene on daphnids occurred much faster in the simulations than in the experiments, increasing

Chapter 4

64

the amount of algae available to rotifers. Also, some daphnid populations were completely eliminated

by pyrene, which resulted in improved B. calyciflorus growth conditions i.e. without competition for

food. In fact, if a similar pyrene effect had occurred on D. magna in the experiments as in the

simulations, an increase in B. calyciflorus in the experiments would be expected.

In general, population dynamics were accurately predicted using a DEBkiss IBM and the DEBkiss IBM

approach is especially applicable for organisms with short life cycles. Despite the limitations of the

DEBkiss IBM to capture the observed population dynamics and structure of D. magna exactly, the

endpoints that are most focused on in traditional ERA – population growth, total population density and

reproduction – were reasonably well predicted. Also, the toxicity models considered here do not account

for size-dependent effects, which limits the impact of a faulty population structure on the population

dynamics after toxic stress.

4.4.2. Applicability of the DEBkiss IBM framework

Without conducting the experiments often deemed necessary for DEB model parameterization (Jager et

al. 2013), the population dynamics of two competing species were simulated reasonably well using

available literature data and universal parameter values (Table 4.1). DEB parameters are often difficult

to measure and interpret. When not available, these were derived from more easily available parameters

using simple equations. For example, the area-specific searching rate of D. magna was derived from the

half-saturation constant for ingestion using Equation (3.4). Although the use of these simple formulas

entails acceptance of certain assumptions (Jager et al. 2013) - e.g. for the example above we assume that

the functional response is a Holling type II - they also offer an easy way to calculate parameter values

that are difficult to determine. Therefore, despite its limitations, DEBkiss IBMs can be useful to predict

population-level effects for species where the large data requirements of typical DEB models is

problematic. Given the limited complexity of DEBkiss IBMs and the problems encountered with

simulating the size structure of the relatively simple D. magna in the current study, their usefulness for

predicting population dynamics of more complex organisms such as fish or large macro-invertebrates

with fluctuating food conditions seems limited. However, they might be used to simulate key life

processes for these organisms such as growth in well-described situations e.g. when food is abundant,

as was done for the pond snail Limnae stagnalis (Jager et al. 2013).

Calibration of a model to only one dataset is not without risk. The validity of the model is only tested

for a specific situation and the parameters can be adjusted to fit the observations without necessarily

capturing the underlying processes(Forbes et al. 2008; Grimm and Martin 2013). In this chapter, the

DEBkiss IBMs were calibrated using one experiment in the case of B. calyciflorus and two experiments

in the case of D. magna. These calibrated models were then applied to three different treatments: pyrene


65

stress, competition and a combination of both. Although calibrated on a limited dataset, these models

were thus applied successfully to very different situations, increasing their acceptability and credibility.

In conclusion, the tested DEBkiss IBMs were able to capture the general patterns of the population

dynamics observed in experiments with D. magna and B. calyciflorus, especially for B. calyciflorus.

Two areas in particular need improvement: the simulated population structure of D. magna and the

toxicity model of pyrene. Improving the modelling of the population structure might be a challenge

because of the absence of a reserve compartment in DEBkiss theory. The toxicity of pyrene could be

better captured using more complex toxicity models that consider other modes of action. This

information is, however, currently lacking for pyrene. For chemicals that have clear acute effects on

survival, the TKTD-SD model implemented here might already prove sufficient.

67

5 DEVELOPMENT OF THE INTEGRATED

CHIMERA MODEL

Chapter 5

68

Abstract

Ecological risk assessment does not aim to protect one or two species in the laboratory but to protect

realistic communities in the field. Two of the most important interactions in a food web are competition

and predation. In this chapter, a DEBkiss IBM implementation for predation was developed and

combined with the earlier developed competition implementation to form the ChimERAfoodweb model. The

ChimERAfoodweb model included two grazers (Brachionus and Daphnia), their predator (Chaoborus) and

two detritus feeders (Asellus and Gammarus). To perform simulations for realistic environmental

conditions, ChimERAfoodweb was coupled with ChimERAfate to form the integrated ChimERA model.

ChimERAfate is a dynamic and spatially-explicit fate model that predicts environmental concentrations

based on environmental variables (hydrodynamics, temperature and trophic state). The integrated

ChimERA model is able to provide spatially-explicit predictions of food web dynamics and the effects

of chemical exposure over time, based on the environmental conditions provided.

Development of the ChimERA model

69

5.1. Introduction

Ecological risk assessment (ERA) procedures were developed to quantify the risk that a chemical poses

in the environment. Ecological risk assessment is typically divided in environmental exposure

assessment and assessment of the potential ecological effects, with distinct procedures for exposure and

effect assessment. Current ERA methods have however received considerable criticism for being

unrealistic, both on the exposure and effect assessment side (Forbes et al. 2008; SCHER (Scientific

Committee on Health and Environmental Risks) et al. 2013; De Laender et al. 2014a). Effect assessment

procedures have been criticized for neglecting important ecological processes such as species

interactions (Fleeger et al. 2003; De Laender et al. 2014a) and spatio-temporal variation in effects of

chemicals (Beketov and Liess 2012). Exposure assessment procedures need to be improved to account

more explicitly for variability in time and space (De Laender et al. 2014a).

Interactions between species, such as competition and predation, can influence the outcome of chemical

stress at the population level by increasing or decreasing the effect of the chemical. For example,

competition with Culex larvae increased the recovery time of Daphnia magna after exposure to

fenvalerate (Foit et al. 2012). Alternatively, in Chapter 2 of this thesis, the effects of pyrene were found

to be lower in D. magna populations experiencing predation because of a higher proportion of large,

more tolerant individuals. Accurately predicting the effects of chemical exposure on higher levels of

biological organization can thus not be done without taking these interactions into account. This has

been recognized in many opinion and review papers and in (regulatory) advisory documents (Fleeger et

al. 2003; Forbes et al. 2009a; SCENIHR et al. 2012).

The landscape structure, the location of exposure within the landscape and the presence of unexposed

populations in the vicinity are other factors determining chemical effects and these are especially

important for the recovery of the affected populations. For example, isolated communities recovered

more slowly from the application of endosulfan than less isolated communities (Trekels et al. 2011). In

another study, the presence of uncontaminated patches increased the recovery time of communities after

lufenuron exposure in mesocosms (Brock et al. 2009). Similarly, the timing of exposure can greatly

influence population dynamics (Relyea and Hoverman 2006). Early life stages, especially embryonic

stages, are often more sensitive than “older” individuals to chemicals. Exposure during periods of

reproduction can thus lead to larger population effects compared to exposure during periods without

reproduction (Bridges 2000; Galic et al. 2012). Also, the exposure pattern of the chemical is important

i.e. where the exposure occurs, how frequently and in what concentrations. Exposure patterns can differ

greatly between chemicals and this leads to different effects on population dynamics (De Laender et al.

2014a). For example, pesticides are typically applied during brief periods of the year. When pesticides

have a rapid degradation rate, species will be exposed for only a short period to a peak concentration.

Chapter 5

70

Pesticides that are more persistent will stay in the environment longer and species will thus be exposed

for a longer time. Other chemicals may have no distinct application time and emission to the

environment will hence be more constant (e.g. industrial chemicals or personal care products).

Many factors, both abiotic and biotic, can thus influence the outcome of exposure to a chemical and it

is impossible to account for all these factors in experiments. Models are a good alternatives and have

been successfully applied in both exposure and effect studies. The dynamic and spatially-explicit

ChimERAfate model, for example, has been developed to predict the chemical concentration in the

different environmental compartments of a shallow pond (Morselli et al. 2015). Similarly, population

models have been used to predict the effects of chemicals on populations. The effect of the insecticide

modelmethrin to populations of three different arthropods (Chaoborus crystallinus, Daphnia magna and

Gammarus pulex) was modelled using individual based models and a mechanistic effect model

(Dohmen et al. 2015). In another study, the influence of where and when a stress event occurred on the

recovery of the isopod Asellus aquaticus was studied using an individual based model (Galic et al. 2012).

Ecological models have typically been applied to predict the effects of chemicals on populations of one

species and thus neglecting possible effects of interspecies interactions. Considering that ecological

models have been suggested as good tools to extrapolate individual-level effects to population, food web

and/or ecosystem level effects (Grimm et al. 2009; De Laender et al. 2014a), it is important to

incorporate interactions between species in these models. One recent modelling study showed for

example that adding interspecific competition to individual based models increased recovery times after

chemical exposure up to three times (Kattwinkel and Liess 2013).

In Chapters 3 and 4, we developed and successfully applied DEBkiss IBMs for two grazers (D. magna

and B. calyciflorus) competing for a shared food source and successfully simulated the effects of a

chemical (pyrene) on the population dynamics of these two species. However, a second key interaction

between species was not yet implemented within the DEBkiss IBM framework: the predator-prey

relationship. Additionally, the DEBkiss IBMs were applied only for populations in a lab environment

which lacked a spatial structure. As mentioned above, the spatial and temporal dimension of chemical

exposure is, however, essential to perform a realistic ecological risk assessment. Also, the two

components of ecological risk assessment – exposure and effect assessment – are closely linked and

interact. For example, detritus and phytoplankton are two key aspects of food webs that also affect the

bio-availability of a chemical (Morselli et al. 2015). In the models presented in the previous chapters we

have neglected the role of environmental fate on the subsequent effects.

The first objective of this chapter was therefore to develop a DEBkiss IBM implementation for predation

and, by combining this with the earlier implementation of competition, to develop a complete food web

model. The second objective was to implement an environmental fate model to predict where and when

exposure will occur. Finally, both submodels were combined to form the ChimERA model, i.e. a


71

spatially-explicit model that can predict the dynamics of a food web exposed to the realistic emission

pattern of a chemical.

5.2. General architecture

The ChimERA model consists of two submodels: the ChimERAfoodweb model and the ChimERAfate

model (Figure 5.1). The ChimERAfoodweb model is used to predict the densities of the different species

in a specified food web and the possible effects of chemical(s) on these densities. The ChimERAfate

model predicts, based on the emission pattern of a chemical and environmental variables (temperature,

trophic state and the hydrodynamics of the system), the concentration of a chemical in the environment

as well as the phytoplankton and detritus concentration. Both submodels exchange information during

simulations: the ChimERAfate model provides the (chemical) exposure concentration, phytoplankton

concentration and detritus concentration to the ChimERAfoodweb model. The ChimERAfoodweb model

provides the amount of phytoplankton and detritus biomass consumed by the food web to the

ChimERAfate model. The ChimERA model is a dynamic spatially-explicit model: chemical fate and food

web dynamics take into account the spatial structure of the system and can provide predictions for

environmental variables that are dynamic over time.

Figure 5.1: Overview of the integrated ChimERA model: based on the trophic state, the hydrodynamics, and the temperature of a system and the chemical exposure present in a system, the integrated ChimERA model provides predictions of the chemical fate and food web dynamics.

5.3. ChimERAfoodweb model

A simple pond food web consisting of five species was considered in the current food web model: two

grazers predated by a predator and two detritus feeders (Figure 5.2). Model genera for the two grazers

were Brachionus and Daphnia, Chaoborus was selected for the predator and Asellus and Gammarus

were used as the two detritus feeders. All five species were modelled using DEBkiss IBMs (Chapter 3).

The advantages of these DEBkiss IBMs over other IBM approaches were their transparency and

simplicity, which allowed easier interpretation of the simulations compared to more complex DEB

Chapter 5

72

approaches. Also, the limited complexity allowed for faster calculations. Parameterization of the

DEBkiss IBMs was done using the ‘add my pet’ database

(http://www.bio.vu.nl/thb/deb/deblab/add_my_pet) and literature sources (Table 5.1). Certain DEBkiss

parameters, mainly feeding-related (maximum specific feeding and searching rate), were adjusted to

allow stable population dynamics using the food levels provided by the fate model. The applicability of

DEBkiss IBMs has not been tested for Asellus and Gammarus. However, the parameters used were

based on the validated ‘add my pet’ database and the application of DEBkiss IBMs for competition has

been tested in Chapter 4. The competition between Asellus and Gammarus is therefore expected to be

realistically captured by DEBkiss IBMs.

Figure 5.2: Configuration of the pond food web used in the current chapter. Full arrows indicate the mass fluxes between the different food web components.

Chemical effects on survival were modelled using a toxicokinetic toxicodynamic (TKTD) approach,

based on the generic universal threshold model for survival (GUTS) assuming stochastic death (Jager et

al. 2011). A detailed description of the DEBkiss IBM and GUTS implementation is given in Chapter 3.

Exposure to a mixture of different chemicals was also implemented in the ChimERA model. Different

mixture toxicity models exist (e.g. concentration addition or independent action) and these are mostly

based on the mode of action of the chemicals concerned (Jonker et al. 2005).


73

Gam

mar

us

4.65

E-02

3.59

E+00

1.91

E-01

8.55

E-05

1.39

E-04

1.00

E-01

8.00

E-01

8.00

E-01

9.50

E-01

8.00

E-01

4.80

E-01

9.00

E-01

4.00

E-01

3.65

E02

1.00

E00

1.00

E-03

4.00

E01

Dap

hnia

1.06

E-02

4.64

E-01

4.08

E-03

1.19

E-03

5.40

E-03

1.00

E-01

8.00

E-01

8.00

E-01

9.50

E-01

8.00

E-01

3.80

E-01

8.00

E-01

4.00

E-01

9.60

E01

1.00

E00

1.00

E-02

2.8E

00

Cha

obor

us

1.15

E-02

6.50

E-02

8.65

E-03

5.00

E-04

2.60

E-03

1.00

E-01

8.00

E-01

8.00

E-01

9.50

E-01

8.00

E-01

1.37

E-01

9.00

E-01

6.00

E-01

1.60

E02

1.00

E00

1.00

E-03

9.00

E01

Bra

chio

nus

2.35

E-02

1.96

E+00

1.40

E-03

3.40

E-02

1.08

E+00

1.00

E-01

8.00

E-01

8.00

E-01

9.50

E-01

8.00

E-01

5.10

E-01

6.00

E-01

8.00

E-01

5.00

E01

1.00

E-01

2.00

E-02

5.00

E-01

Ase

llus

5.32

E-02

3.25

E+00

4.50

E-03

9.96

E-03

5.21

E-01

1.00

E-01

8.00

E-01

8.00

E-01

9.50

E-01

8.00

E-01

4.20

E-01

9.00

E-01

4.00

E-01

3.65

E02

1.00

E00

1.00

E-03

8.00

E01

Para

met

er

Max

imum

spec

ific

assi

mila

tion

rate

(m

g dw

mm

-2 d

-1)

Spec

ific

sear

chin

g ra

te (L

mm

-2 d

-1)

Vol

ume-

spec

ific

mai

nten

ance

cos

ts (m

g dw

mm

-3 d

-1)

Ass

imila

tes i

n a

sing

le fr

eshl

y la

id e

gg (m

g dw

)

Stru

ctur

al b

ody

mas

s at p

uber

ty (m

g dw

)

Dry

-wei

ght d

ensi

ty o

f stru

ctur

e (m

g m

m-3

)

Yie

ld o

f stru

ctur

e on

ass

imila

tes (

grow

th,,

-)

Yie

ld o

f ass

imila

tes o

n st

ruct

ure

(sta

rvat

ion,

-)

Yie

ld fo

r con

vers

ion

of re

prod

uctio

n bu

ffer

to e

ggs

(-)

Yie

ld o

f ass

imila

tes o

n fo

od (-

)

Shap

e co

rrec

tion

coef

ficie

nt (-

)

Frac

tion

of a

ssim

ilate

s for

gro

wth

and

mai

nten

ance

(-)

Bod

y m

ass f

ract

ion

at w

hich

star

vatio

n oc

curs

(-)

Max

imum

age

of a

n in

divi

dual

(d)

Mov

emen

t pro

babi

lity

(d-1

)

Stoc

hast

ic m

orta

lity

rate

(d-1

)

Tim

e be

twee

n re

prod

uctio

n ev

ents

(d)

Sym

bol - - - - -

Tabl

e 5.

1: D

EBki

ss IB

M p

aram

eter

val

ues

used

for

the

five

spec

ies

cons

ider

ed in

the

Chi

mER

Afoo

dweb

mod

el. P

aram

eter

val

ues

wer

e ta

ken

from

the

‘add

my

pet’

data

base

(http

://w

ww

.bio

.vu.

nl/th

b/de

b/de

blab

/add

_my_

pet)

and

from

lite

ratu

re so

urce

s (Ap

pend

ix C

).

Chapter 5

74

Often, a priori knowledge on how chemicals interact is not available and an informed choice of the

correct mixture toxicity model is therefore not possible. Hence, as an alternative approach, the hazard

of each chemical was calculated (see equations 3.14-3.16) and summed, assuming that the hazard of a

chemical was not influenced by other chemicals. The summed hazard was then used to calculate the

survival over time using

(5.3)

with S(t) the survival probability of an individual at time t (d); HZ the cumulative hazard caused by

chemical i of an individual at time t (Equation 3.16) and n the number of chemicals to which the

individual is exposed.

Some species-specific adjustments were done to enhance the realism of the DEBkiss IBMs. For Asellus,

Daphnia and Gammarus, i.e. species with carapaces and appendices which are unable to shrink, the

feeding flux was determined using their maximum attained structural length. All species were assumed

to follow the general DEBkiss life cycle: individuals start their life cycle as an embryo, hatch as juveniles

and grow to the adult life stage, where reproduction can occur. This is roughly applicable for all species

except for Chaoborus. The Chaoborus present in the food web are actually the larval stage and

emergence of the free-flying stage is needed for reproduction (Van Wijngaarden et al. 2006). This latter,

free-flying life stage was ignored in the DEBkiss IBM implementation because its duration is small

compared to that of the waterborne life stages (Von Ende 1982). So, when reproduction occurred, the

Chaoborus individual was removed from the population and its energy available in the reproduction

buffer was converted into eggs.

Competition between grazers and detritus feeders was modelled using resource competition only, as

described in Chapters 3 and 4 for experiments with Brachionus and Daphnia. With each time step, the

biomass consumed by each species was summed and, if the theoretical consumption was higher than the

available food, rescaled to the available biomass. Although density-dependent effects on mortality can

be expected for e.g. the detritivores Asellus and Gammarus (Galic et al. 2012; Dohmen et al. 2015), they

were not considered in our hypothetical scenarios. The outcome of competition was thus solely

determined by the rate and the efficiency of food uptake and the ability to cope with low food conditions.

Predation on the two grazers by Chaoborus was modelled following a multi-species predation approach

(Krylov 1992; Rose et al. 1999):

with ′

′ (5.4)

with the biomass of prey j eaten by predator i (mg dw d-1); the maximum ingestion by the

predator (mg dw d-1); the weight of predator i; f the functional response parameter (-); the


75

handling time of prey j (d); ′ the attack rate on prey j (L d-1); the prey density of prey j; and n the

amount of prey species. A detailed description of the derivation of the predation equations can be found

in Appendix C. The used attack rate and handling time were 0.2 L d-1 and 0.76 d for Brachionus (Krylov

1992), respectively and 0.11 L d-1 and 0.07 d for Daphnia (Moore 1988), respectively. Similar to the

adjustment of the feeding-related DEBkiss parameters, these attack rates and handling times were

adjusted literature values to avoid overexploitation of the prey species.

The landscape implemented in the ChimERA model consisted of two connected ponds and inflow and

outflow streams. The spatial landscape was implemented in the ChimERAfoodweb model by dividing the

two pond system into 92 equal-sized patches of 10 m² (Figure 5.3). All patches have individual

phytoplankton, detritus and population dynamics. Movement between patches was implemented as

random movement: a species-specific movement parameter defined the probability to move to a

neighbouring water patch (Table 5.1). All species except Brachionus were assumed to move on average

1 patch per day. Brachionus movement was considered lower and individuals only moved 0.1 patch per

day, on average.

Figure 5.3: Spatial discretization of the two-pond system in the ChimERAfoodweb model. The spatial landscape is divided in 92 patches of 10m².

5.4. ChimERAfate model

Because the focus of this PhD thesis is on the ecological effects of chemical exposure, only a brief

description of the chemical fate model is given here. For a detailed description of the ChimERAfate

model, see Morselli et al. 2015 (Morselli et al. 2015). In short, the ChimERAfate model was based on the

fugacity approach (Mackay 2001) and modelled the water-sediment system in two connected 450 m²

ponds (Figure 5.4). Pond depth and sediment layer depth were assumed to be 50 cm and 1 cm,

respectively. Phytoplankton was assumed to be the dominant autotrophic group in the water column and

macrophytes were therefore not considered as a compartment. Different compartments and sub-

compartments were taken into account to predict the concentration of a chemical: water (water and

Chapter 5

76

suspended solids), sediment (pore water and sediment solids), dissolved organic carbon, particulate

organic carbon and phytoplankton. Spatial discretization was obtained by dividing the two pond system

in 20 slices and describing the one-dimensional water flow between these slices. A hydrological module

was used to compute water volumes (m3) and fluxes (m3 h−1) on an hourly basis in the slices. Slices

reflect areas with identical hydrodynamic properties but are not necessarily equal in area (Figure 5.4).

All calculations in the fate model were performed for each of these 20 slices.

Because the model was dynamic, time-varying environmental variables could be used as an input for

the model. Important environmental variables included the temperature, hydrodynamics and trophic

state of the system. Temperature is not constant throughout the year and varies significantly between

regions. The temperature influences partitioning constants and rates in the fate model e.g. the Henry law

constant and the dissolution rate of detritus. Trophic state refers to the nutrient conditions in the system

and is often described qualitatively as either oligotrophic, mesotrophic or eutrophic. The nutrient

conditions will determine the phytoplankton concentrations in the system. The hydrodynamics of the

systems describe the water flow. An often used measure for the water flow is the water residence time:

the time needed for all water in a location to be replaced. The model was also capable of using time-

varying chemical emissions as input to predict chemical concentrations in the different compartments at

different time points.

Figure 5.4: The spatial configuration in the ChimERAfate model. (A) Side view of the two connected ponds. Blue arrows indicate the direction of the water flow, red arrows represent the flow of the chemical. (B) Top-down view of the two connected ponds. Red dashed lines indicate the division of the system in 20 slices by the ChimERAfate model.


77

The compartments included in the ChimERAfate model that were relevant for the ChimERAfoodweb model

were the phytoplankton, detritus and water concentration of the chemical. Phytoplankton and detritus

dynamics were described in the ChimERAfate model using ordinary differential equations (Equations 5.1

and 5.2). The phytoplankton and detritus concentrations determined the energy available for the

individuals in the food web. The food web model in turn provided the phytoplankton and detritus losses

due to consumption. The water concentration of the chemical was used to determine possible effects on

the food web.

Phytoplankton growth was not constant throughout the year and was described by

(5.1)

with the change in phytoplankton concentration over time (mg ww d-1 L-1); the phytoplankton

concentration (mg ww L-1); the time (days); the gross primary production (d-1); the fraction

of spend on respiration (-); the fraction of spend on excretion (-); the carrying capacity

(mg ww L-1); the mortality rate of phytoplankton (d-1) and the phytoplankton losses due to

grazing (mg ww L-1 d-1).

Detritus concentration was closely linked to the phytoplankton dynamics and was described by

(5.2)

with the change in detritus concentration over time (mg ww d-1 L-1); the phytoplankton

mortality rate (d-1); the phytoplankton concentration (mg ww L-1); the detritus dissolution rate

(d-1); the detritus concentration (mg ww L-1) and the detritus losses due to consumption by

detritus feeders (mg ww d-1 L-1).

Gpp and K were two parameters that are specific for a given situation and were chosen on a case by case

basis. Standard values were taken for the other phytoplankton and detritus model parameters (De

Laender et al. 2015): Resp = 0.1; Excr = 0.1; Mort = 0.2 d-1; Dis = 0.01 d-1.

5.5. Technical implementation

The main technical challenge was coupling the fate model and the food web model which differed

greatly in their architecture and internal communication. Instead of implementing direct communication

between the two models, the ChimERAfate model and the ChimERAfoodweb model were coupled by

coordinating the communication via a server (Figure 5.5). The advantage of this approach is that it is

very flexible i.e. different models can easily be plugged in, regardless of the encoding language since

only simple text strings with the necessary information are exchanged with the server. As long as they

Chapter 5

78

can communicate with the server, new models can be connected to the server. This makes it easy to e.g.

switch to a different food web model without needing to recode the server or fate model.

A detailed description of how the communication between the two models was achieved is given in the

communication protocol provided in Appendix C. In short, at each time step both the fate client and the

food web client send the requested data to the server. The server then selects which data are needed by

individual clients and forwards it to the respective clients. The server then asks the connected models to

proceed to the next time step using the provided information. For example, at the start of a time step,

the server requests the phytoplankton concentrations in all slices from the fate model. The fate model

sends the requested data to the server. The server then sends these phytoplankton concentrations to the

food web model and asks for the grazing losses. The food web model uses the phytoplankton

concentrations to calculate the biomass of phytoplankton grazed in each slice. These grazing losses are

then sent to the server, which communicates it back to the fate client to calculate the phytoplankton

concentrations for the next time step.

The server was developed using Pharo v4.0 (http://pharo.org). The ChimERAfate model was developed

in Microsoft Visual Basic 6.0. DEBkiss IBMs were available in NetLogo (Wilensky 1999), as described

in Chapter 3. To limit the communication necessary between server, fate model and NetLogo models, a

Java framework was developed. This Java framework calls a NetLogo client per species and summarizes

the information that needs to be exchanged e.g. the grazing losses provided by the Daphnia and

Brachionus NetLogo clients are summed by the Java framework before being communicated to the

server.

Figure 5.5: Technical implementation of the ChimERA model. A server coordinates the communication between the different models. A client translates the server request and commands for the individual models. The food web model is divided in two parts, a detritus-based part (with ASE = Asellus and GAM = Gammarus) and a phytoplankton based part (with BRA = Brachionus, CHA = Chaoborus and DAP = Daphnia).


79

Phytoplankton and detritus concentrations at the start of the year are too low to support the food web.

In reality, species have different strategies to overwinter. Therefore, we assumed that food web

dynamics only started when a critical food concentration of 0.025 mg ww L-1 was reached. Simulations

were started with 100 ind L-1. Early simulations showed that populations of Daphnia and Brachionus

can reach high densities (> 106 individuals in the whole system) which increased the simulation times

considerably. For the two grazers, a superindividual approach was therefore implemented to limit the

computation times. Following the kiss principle, an implementation with limited complexity was

chosen. Superindividuals were considered as one constant unit and were simulated in the same way as

normal individuals (Grimm and Railsback 2013). The only adjustment made was that the amount of

food eaten by the grazers was scaled to the superindividual level i.e. 100 individuals per superindividual

for Daphnia and 1000 individuals per superindividual for Brachionus. Although this approach limits the

variability in a population, it requires few additional assumptions and allows for an easy interpretation

of the simulations.

5.6. Applications

The ChimERA model developed here is an innovative method because it immediately links realistic

exposure predictions of chemicals to effects on food web dynamics. The flexibility and modularity of

the model make it possible to apply the ChimERA model to any environmental and ecological scenario.

Scenarios can differ, among others, in their spatial structure, identity of the emitted chemical, emission

strength and frequency, species in the food web and food web structure. Scenarios can be simulated by

supplying the currently used submodels with the necessary information or, if the currently implemented

submodels are insufficient, by plugging in a more suitable fate or food web model. To demonstrate its

potential, the use of the ChimERA model as a scenario analysis tool is explored in the next chapter. The

ChimERA model will be used to study how differences in the environmental variables of a scenario lead

to differences in the effects of the applied chemicals on food web dynamics.

81

6 THE CHIMERA MODEL AS A

SCENARIO ANALYSIS TOOL

Chapter 6

82

Abstract

The integrated ChimERA model was developed as a tool to perform more realistic ecological risk

assessments. To test its performance, the ChimERA model was applied to a two-pond system connected

by a stream in which 15 hypothetical scenarios – differing in water residence time, temperature, trophic

state and applied chemical (carbendazim, chlorpyrifos, pyrene and the mixture of the three chemicals)

– were evaluated. The predicted effects of the applied chemicals were clearly different for the selected

scenarios. Densities of species were generally highest in scenarios with a high trophic state temperature

and water residence time. Daphnids were the dominant grazers in the ponds while rotifers dominated

in the streams. Concentrations of applied chemicals were highest in scenarios with a high water

residence time. The physico-chemical properties of the chemical determined the spatial and temporal

pattern of the exposure i.e. where the concentrations were highest and how fast the chemical

disappeared. As expected from their sensitivities to the chemicals, direct effects were predicted on

Chaoborus, Daphnia and Gammarus. These effects were, however, heterogeneously distributed in space

and time and reflected the predicted differences in exposure. The most notable indirect effects were

shifts in dominance from Daphnia and Gammarus to Brachionus and Asellus, respectively. Also, the

predator Chaoborus was affected indirectly through effects on its prey species Daphnia. For the mixture

of three chemicals, the effects of pyrene dominated for Daphnia and the effects of carbendazim and

chlorpyrifos for Gammarus. These simulations demonstrate how much the outcome of chemical

exposure is determined by environmental conditions, which cannot be captured with traditional risk

assessment methods. Modelling tools like the ChimERA model can prove essential to answer the current

challenges of ecological risk assessments.

The ChimERA model as a model analysis tool

83

6.1. Introduction

The ChimERA model was developed (Chapter 5) to demonstrate (proof of principle) how available

knowledge and models can be combined into a predictive risk assessment tool applicable in a range of

environmental conditions (De Laender et al. 2014a). To evaluate and demonstrate how such a type of

model could be used to perform scenario analysis, the ChimERA model was applied to 15 hypothetical

scenarios that differed in their environmental conditions and applied chemical(s). Although the main

focus of this chapter is on the ecology and food web dynamics, a short discussion of the chemical fate

model is also included.

6.2. Methods

6.2.1. Scenario description

Three environmental variables can be manipulated in the ChimERA model: temperature, trophic state

and hydrodynamics. Trophic state refers to the amount of nutrients available in the system: oligotrophic,

mesotrophic or eutrophic. Hydrodynamics refer to how fast the water flows and thus the retention time

of the water in the ponds. As numerous combinations of these environmental variables and thus

environmental scenarios are possible, we selected three different ‘example’ combinations of these

variable to evaluate the potential applications and limitations of the ChimERA model (Table 6.1):

Scenario I is a high velocity system with a low average annual temperature and an oligotrophic nutrient

status. Scenario II is characterized by intermediate water velocity, medium average annual temperature

and a mesotrophic nutrient status. Scenario III has a low water velocity, high average annual temperature

and eutrophic nutrient conditions.

Table 6.1: Description of the selected scenarios. Scenarios differ in their temperature, trophic state and hydrodynamics. Annual minimum and maximum temperatures are shown. K = phytoplankton carrying capacity; Gpp = gross primary production; RT = residence time.

Scenario Temperature (Di Guardo et al.)

Trophic State (Håkanson and Peters 1995)

Hydrodynamics (Peretyatko et al. 2007)

I [4.2 - 20.1] °C K = 0.3 mg ww L-1; Gpp = 0.36 d-1

RT = 5 d

II [4.3 - 22.0] °C K = 0.5 mg ww L-1: Gpp = 0.42 d-1

RT = 25 d

III [6.9 - 26.3] °C K = 1 mg ww L-1: Gpp = 0.71 d-1

RT = 100 d

Chapter 6

84

These three selected environmental scenarios were combined with five exposure scenarios: no exposure,

exposure to carbendazim, exposure to chlorpyrifos, exposure to pyrene and finally exposure to all three

chemicals simultaneously. The three selected chemicals differ greatly in their physico-chemical

properties and their exposure pattern (Appendix D: Table D1). Carbendazim is a fungicide used to

control plant diseases (van Wijngaarden et al. 1998) and chlopyrifos is a widely-used insecticide (van

der Hoeven and Gerritsen 1997). Pyrene is a polycyclic aromatic hydrocarbon (PAH) that is mostly

formed as a by-product of anthropogenic activities (e.g. fossil fuel combustion). Pyrene mainly enters

the aquatic environment accidentally through e.g. petroleum spills, wastewater or surface run-off

(Nikkilä et al. 1999). Actual emission rates for these chemicals are difficult to find and are highly site-

specific. The current study is a proof of principle assessment and the performance of the ChimERA

model can best be discussed when effects are expected to occur. Therefore, emission strengths were

chosen to result in maximum water concentrations close to the acute LC50-values of the most sensitive

species in the intermediate scenario II (Appendix D: Table D2). Both pesticides were considered to have

a highly seasonal emission pattern, with short application periods in spring (day 180) and late summer

(day 260; Figure 6.1). The length of the application periods was ten days: emission concentrations

increased during the first five days and then decreased again. Maximum emission strengths were 3.7 ∙

10-4 mol L-1 h-1 and 6.7 ∙ 10-7 mol L-1 h-1 for carbendazim and chlorpyrifos, respectively. Pyrene was

considered to have a continuous ambient exposure pattern, with stable emission concentrations of 2 ∙ 10-

6 mol L-1 h-1 throughout the year (Figure 6.1).

The five species in the food web (Chapter 5) differed in their sensitivity to these three chemicals

(Appendix D: Table D2). Gammarus and Daphnia were most sensitive to carbendazim: the acute LC50

was 55 μg L-1 and 91 μg L-1, respectively (van Wijngaarden et al. 1998). No effects on Chaoborus were

observed in toxicity tests at concentration of up to 3435 μg L-1 (van Wijngaarden et al. 1998). For

chlorpyrifos, Gammarus, Chaoborus and Daphnia were considered the only sensitive species: the acute

LC50 was 0.23 μg L-1, 0.3 μg L-1 and 0.8 μg L-1, respectively. Gammarus and Daphna were also the most

sensitive of the five species when exposed to pyrene: the acute LC50 was 27.1 μg L-1 and 68 μg L-1,

respectively (Chapter 2). No pyrene effects were observed at nominal concentrations of up to 2000 μg

L-1 for Brachionus and Chaoborus (Chapter 2). Brachionus toxicity test results were not available for

carbendazim and chlorpyrifos but rotifers are often more tolerant to chemicals than cladocerans (Girling

et al. 2000) and no effects of these chemicals on rotifers were therefore assumed at the used exposure

concentrations.

Because mortality and organism mobility in the DEBkiss IBMs are stochastic processes, variability is

expected between simulations. To account for this variability, simulations are typically iterated multiple

times. However, in this case, simulation times for one scenario were as high as 3 hours, limiting the

number of iterations possible. As a compromise, each scenario was run five times. Ideally, model

performance is assessed by comparing model simulation with observations. However, these


85

observations were not available for the system and exposure scenarios modelled here. As an alternative

approach, the scenario expectations were defined based on a priori knowledge and compared with the

simulations. The expected environmental fate of each chemical was defined based on the physico-

chemical characteristics,. For the food web, the interactions between species and the sensitivity of each

species to the chemical were taken into account to define a priori expected population dynamics.

Figure 6.1: Used emission patterns for (A) carbendazim, (B) chlorpyrifos and (C) pyrene.

6.2.2. Expectations regarding the outcome of the scenario analyses

6.2.2.1. ChimERAfate

In this section, a short description is given for each chemical of how their physico-chemical properties

(Appendix D: Table D1) influence their environmental fate and how this is altered by changes in

temperature, trophic state and water retention time. At the end of this section, a summary is given of the

expectations for each chemical per scenario.

Chapter 6

86

Carbendazim is a hydrophilic substance (logKow = 1.52) and will thus not accumulate in the sediment.

Moreover, degradation is slow (HLwater = 720 h). Carbendazim will thus mainly disappear from the

system with the water outflow and, because of the low sediment accumulation, the short peaks in

emission are expected to be reflected by short peaks in water concentrations. The water retention time

will thus be an important factor for its environmental fate. Chemical concentrations will be higher in

systems with high water retention time: for the same amount of chemical emitted over an hour, the

chemical is diluted less when the water is flowing slowly. Carbendazim will thus be present in the water

longer and in higher concentrations in systems with high water residence times. Because of the limited

sediment accumulation and degradation, carbendazim concentrations are not expected to be influenced

much by temperature or the trophic state of the system, which mainly influence degradation and

accumulation.

Chlorpyrifos is a more hydrophobic substance (logKow = 4.96) and degrades quickly (HLwater = 24 h).

Accumulation in the sediment and degradation are thus important environmental processes for this

chemical. As a result of the high sediment accumulation, a long period of reduced but constant water

concentrations are expected after a peak in emission. More organic material is available in eutrophic

systems compared to oligotrophic systems, which leads to, among others, more accumulation in the

sediment. Temperature increases the phytoplankton growth and biomass, further increasing the amount

of organic material available in the system. Also, the temperature positively influences the degradation

rate of the chemical. Temperature and trophic state will thus determine the water concentrations together

with the water retention time. As described for carbendazim, higher water concentrations are expected

for systems with high water retention time. However, high water retention times also increase the time

available for accumulation to the sediment and for degradation, both decreasing the water concentration

of the chemical. Which process is dominant is hard to predict a priori because this depends on the

relative strength of the environmental fate processes.

Pyrene is a hydrophobic substance (logKow = 5.18) that degrades slowly (HLwater = 17 ∙ 103 h). Pyrene

will thus accumulate in the sediment and persist. Because of the continuous emission pattern, constant

water concentrations are expected once equilibrium between all environmental compartments is

reached. As was the case for chlorpyrifos, temperature and trophic state are expected to increase the

accumulation in the system, decreasing the water concentrations of pyrene. The effect of water retention

time is more predictable than for chlorpyrifos: a high water retention time will increase the water

concentrations because of the lower dilution of the chemical. Effects of the water retention time on

accumulation and degradation are negligible because of the continuous emission of pyrene.

In summary, for the three selected environmental scenarios, carbendazim concentrations in scenario III

are expected to be higher than in scenario II and I because of the longer residence times (100 days versus

25 and 5 days, respectively; Table 6.1). Peaks in carbendazim concentration are expected to disappear


87

quickly in all scenarios because of the limited accumulation, with the narrowest peak in scenario I. For

chlorpyrifos and pyrene, it is harder to predict a priori how the concentrations will differ between the

scenarios. Water concentrations are expected to decrease with temperature and trophic state but the

opposite is expected for water retention time. The differences between scenario I, II and III will thus

reflect the relative importance of these three environmental variables. Chlorpyrifos peaks are expected

to persist longer in scenario III than in scenario II and I because of the longer residence time and, as a

result of the higher temperature and trophic state, the increased accumulation in the sediment. As was

the case for chlorpyrifos, differences in pyrene concentrations between the three scenarios are hard to

predict. Pyrene concentrations are expected to become constant once equilibrium between the

compartments has been reached.

6.2.2.2. ChimERAfoodweb

This section starts with a description of how the food web dynamics are expected to be influenced by

the environmental conditions. Next, the expected effects of the applied chemical(s) on the food web

dynamics are discussed. Finally, a summary of the expected outcomes of the scenarios is given.

The phytoplankton carrying capacity and gross primary production, indicators for the trophic state in

the fate model, will determine the phytoplankton biomass and growth. A higher phytoplankton biomass

means more energy is available for the grazers. The densities of the grazers Daphnia and Brachionus

and their predator Chaoborus are thus expected to be higher when the trophic state of the system is high.

Phytoplankton is the main source for detritus and detritus dynamics are thus closely linked to the

phytoplankton dynamics. In the absence of phytoplankton grazing, a high trophic state is expected to

lead to high detritus concentrations, which allows the densities of the detritus feeders Asellus and

Gammarus to be higher. Phytoplankton grazing will limit the biomass of phytoplankton being converted

to detritus, making it harder to predict how the detritus concentrations will differ between trophic states.

Because of the close link between phytoplankton and detritus, patterns in the phytoplankton dynamics

are expected to be reflected in the detritus dynamics. Because of the strong relation between their food

sources, patterns in grazers and detritus feeders are also expected to be linked. Chaoborus dynamics in

turn are expected to reflect the dynamics of its food source, the grazers. Phytoplankton drift with the

water is not considered in the fate model and the water retention time thus should not influence the food

web dynamics.

Gammarus and Daphnia are the most sensitive species to the three chemicals considered here. Exposure

to these three chemicals is therefore expected to lead to decreased abundances of Gammarus and

Daphnia. Direct effects of chlorpyrifos are also expected for Chaoborus but not for the other two

chemicals. As indirect effects, the densities of the more tolerant competitors Asellus and Brachionus are

expected to increase. Chaoborus feeds on both Daphnia and Brachionus and because both prey species

Chapter 6

88

are differently affected by carbendazim and pyrene, it is difficult to predict how these two chemicals

will affect the predator dynamics. In the case of chlorpyrifos, direct effects are expected for both

Chaoborus and Daphnia. Indirect effects on Brachionus are thus expected to be especially strong

because of both reduced competition and predation by chlorpyrifos. Since chlorpyrifos and carbendazim

are applied only during short periods, (partial) recovery of the Gammarus and Daphnia densities should

be possible after the application periods. Pyrene exposure is constant throughout the year and

consistently different food web dynamics from the control scenarios are therefore expected. When

exposed to the mixture of all three chemicals, the continuous exposure to pyrene is expected to affect

the food web the most because the exposure starts earlier: Daphnia and Gammarus densities will

decrease and indirect effects on Asellus, Brachionus and Chaoborus will occur. The application of the

other two chemicals will further increase the effects of pyrene, leading to higher effects than observed

in the individual exposures.

In summary, the densities of all species in the food web are expected to be higher in scenario III than in

scenario II and I because of the higher trophic state and temperature in that scenario. Direct effects of

the chemicals are mainly expected for Daphnia and Gammarus while Asellus, Brachionus and

Chaoborus will be indirectly affected through reduced competition or reduced prey availability. The

magnitude of the predicted effects of the chemicals on the food web is closely linked to the expected

exposure patterns (See 6.2.2.1). For carbendazim, higher concentrations are expected in scenario III than

in scenario II and I, and higher effects on the food web dynamics are thus also expected. For chlorpyrifos

and pyrene, the expected concentration differences between the scenarios are less clear and it is thus

hard to predict differences in effects between the scenarios.

6.3. Results

For a clear presentation of the simulations, four slices (see 5.4) were chosen that were representative for

the whole system (Figure 6.2): slice 1 was located at the inflow of the first pond i.e. where the chemicals

were emitted into the system, slice 5 was located in the middle of the first pond, slice 10 was located in

the stream between the first and second pond and slice 15 was located in the middle of the second pond.

The observed patterns in the simulations are described in this section for the ChimERAfate (6.3.1) and

ChimERAfoodweb (6.3.2) model and compared to the expected patterns. The evaluation of the underlying

processes is described in section 6.4.


89

Figure 6.2: Location of slices 1, 5, 10 and 15, for which the population dynamics are shown in Figures 6.3-6.13.

6.3.1. ChimERAfate

In this section, the expected results of the ChimERAfate model are compared with the performed

simulations for each chemical. The dominant fate processes for each chemical are also provided. For a

discussion of the observed patterns and the underlying processes, see 6.4.1.

As expected because of the higher water residence time, carbendazim concentrations in slice 1 were

higher in scenario III (971.87 μg L-1) than in scenario II (255.54 μg L-1) and scenario I (51.92 μg L-1).

Because of the increased water retention time, the decrease in carbendazim concentrations throughout

the two-pond system was also higher in scenario III than in scenario II and scenario I (Figure 6.3): the

carbendazim concentration in the outflow stream (slice 20) was 50.9%, 11.1% and 2.5% of the

concentration in the inflow stream (slice 1) for scenario I, II and III, respectively. Because of the limited

sediment accumulation, concentrations of carbendazim are expected to decrease sharply when emission

stops. This was indeed observed for scenario I and scenario II but not for scenario III. In scenario III,

the carbendazim concentration only decreased sharply after emission in slice 1, where emission

occurred. Further downstream, peaks were much less pronounced (slice 5) or not observed (slice 15).

Outflow with water, diffusion to sediment pore water and degradation were the most important fate

processes for carbendazim.

Chlorpyrifos concentrations were higher in scenario III (1.55 μg L-1) than in scenario II (0.42 μg L-1)

and I (0.09 μg L-1), indicating that water retention time was more important than temperature and trophic

state (see 6.2.2.1). As expected, a reduced but constant chlorpyrifos concentration was predicted after

the application period because of the accumulation in the sediment (Figure 6.3). The peak chlorpyrifos

concentrations decreased sharply further downstream and this was more pronounced with higher water

residence times: peak chlorpyrifos concentrations in slice 20 were 21.8%, 0.3% and <0.01% of the peak

Chapter 6

90

concentrations in slice 1 for scenario I, II and III, respectively. Outflow with water and degradation were

the most important processes that determined the chlorpyrifos water concentration.

Pyrene concentrations were higher for scenario III (6.79 μg L-1) than in scenario II (1.73 μg L-1) and

scenario I (0.35 μg L-1) and, as expected, concentrations stayed constant once equilibrium was reached

(Figure 6.3). This equilibrium between all environmental compartments was reached later further away

from the emission source. Pyrene concentrations decreased further downstream and this was more

pronounced in scenarios with a high water retention time: pyrene concentrations in the outflow stream

were 87.0%, 52.5% and 9.9% of the concentrations in the inflow stream for scenario I, II and III,

respectively. For pyrene, the dominating fate processes were the outflow with water and the particle

deposition to the sediment.


91

Figure 6.3: D

issolved water concentrations of the three m

odel contaminants as predicted by the C

himERA m

odel for three environmental

scenarios (scenarios I, II and III) and four selected locations in th e two-pond system

exposed for one year: at the inflow of the first pond

(Slice 1), in the middle of the first pond (Slice 5), betw

een the two ponds (Slice 10) and in the m

iddle of the second pond (Slice 15).

() lll a-<1> g_ lll t:!. 3

() :::r 0

~ 5 "'

"1J '< (i) :::1 <1>

Water con centration (1-19 L -1)

1e-o6

8

8

8

1e-o6

8

8

8

-i 3" (!) 1e-06 -a. 0

8

8

8

1e-o6

1e-Q2

1e-Q2

1e-Q2

1e-Q2

,F\

1e+02

1e+02

1e+02

1e+02 1

8~

!H ) I I

1e-ll6 1e-Q2 10+02

8

:~ 1e-ll6 1e-Q2 1e+02

8

8

8

1e-ll6 1e-Q2 1e+02

8

8

8

1e-ll6 1e-Q2 1e+02

I : l I

· ~ I

I 8 ~ ~ î)

I !H / I )

1e-ll6 1e-Q2 1e+02

8

:b 1e-ll6 1e-Q2 1e+02

8

8

8

1e-ll6 1e-Q2 1e+02

8

8

8

1e-ll6 1e-Q2 1e+02 I I I I

I : l l I 8 ~ r:::::;:-r

I 8 I - ~

tQ r 0 m --"

tQ r 0 m (.Tl

tQ r () m --" 0

ltQ c

I~ --"

I (.Tl

Chapter 6

92

6.3.2. ChimERAfoodweb

In this section, the simulated food web dynamics are described and compared with the expectations.

First, the scenarios with no exposure are discussed, next the carbendazim, chlorpyrifos and pyrene

exposure and finally the exposure to the mixture of all three chemicals. The underlying processes are

discussed in section 6.4.2. and 6.4.3.

6.3.2.1. No exposure

As expected, food web dynamics of the three environmental scenarios (I, II and III) differed greatly

(Figures 6.4 and 6.5). In general, following expectations, population densities increased with an increase

in the trophic state of the scenario e.g. the maximum population density of Brachionus was 101 ∙ 103

ind L-1, 224 ∙ 103 ind L-1 and 1467 ∙ 103 ind L-1 for scenario I, II and III, respectively. This pattern was

influenced by interactions between species: Daphnia was the dominant grazer in scenario I and II,

preventing the Brachionus density to increase. The dominance pattern shifted in the highest trophic level

(scenario III): Brachionus was able to benefit more from the increased phytoplankton concentrations

than Daphnia and became the dominant grazer. Because of the strong competition with Brachionus,

Daphnia densities in scenario III (733 ∙ 102 ind L-1) were not higher than the Daphnia densities in

scenario II (740 ∙ 102 ind L-1). Chaoborus can predate on both grazers. However, Chaoborus densities

were highest in scenario II (403 ind L-1), when Daphnia was dominant and not in scenario III (244 ind

L-1), when Brachionus densities were high. The population dynamics of Chaoborus were closely linked

with the Daphnia dynamics but not so much with the Brachionus dynamics.

The higher population densities of the grazers also led to a more dynamic system in scenario III, with a

typical cyclical pattern of phytoplankton growth, increase in grazer densities, phytoplankton decline and

decrease in grazer densities. This pattern was also visible but with fewer peaks per year in scenario I

and II where Daphnia followed the phytoplankton dynamics. Because of the grazing pressure, the

phytoplankton was not able to grow until the carrying capacity of the system (K): maximum

phytoplankton concentrations were 0.16 mg ww L-1 (K = 0.30 mg ww L-1), 0.31 mg ww L-1 (K = 0.50

mg ww L-1) and 0.79 mg ww L-1 (K = 1.00 mg ww L-1) for scenario I, II and III, respectively. The

detritus concentration reflected the phytoplankton concentration (Figure 6.5): more fluctuations and

higher maximum detritus concentrations in scenario III (0.58 mg ww L-1) than in scenario II (0.22 mg

ww L-1) and I (0.10 mg ww L-1).

For the detritus feeders, Gammarus was the most dominant species in all scenarios while Asellus

densities remained low (Figure 6.5). The highest densities of both detritus feeders were observed in

scenario II, despite the higher detritus concentration in scenario III. Unexpectedly, the population

dynamics of the detritus feeders differed greatly of those of the grazers: the densities of the detritus


93

feeders mainly increased in the second half of the year while periods of population growth were observed

for grazers throughout the year.

Unexpectedly, clear spatial differences were observed between the species: population densities of

Daphnia and Chaoborus were much higher in the ponds (Slice 5 and 15) than in the inflow and

connecting streams (Slice 1 and 10). Based on the Brachionus dynamics in scenario III, spatial

differences are much less prominent for this species. Similar to Daphnia and Chaoborus, Asellus and

Gammarus population densities were highest in the ponds and lower in the inflow and connecting

streams. Contrary to the ponds, the reduced competition with Gammarus in the streams allowed short

periods of population increase for Asellus e.g. slice 1 and 15 in scenario I around day 230.

Chapter 6

94

Figure 6.4: Food web dynamics without chemical exposure as predicted by the ChimERA model for three environmental scenarios (scenarios I, II and III) and four selected locations in the two-pond system: at the inflow of the first pond (Slice 1), in the middle of the first pond (Slice 5), between the two ponds (Slice 10) and in the middle of the second pond (Slice 15). Only the phytoplankton-based part of the food web is shown. Lines and shaded areas indicate the average and minimum and maximum densities, respectively, of five iterations per scenario.

Q) (.) c ro "0 11 c :::J .0 <(

111

SLICE 1 SLICE 5 ~ ~ ~

~ ~ r ~ ~

;j I ~ I

[ ~ ; 100 150 200 250 300 350 100 150 200 250 300 350

lil ~ lil

i ~ ~ ~

~ "' 8 0 V

~ f\/\f'\.._ lQ 0

. 0 0 N

100 150 200 250 300 350 100 150 200 250 300 350

~ - ~ i!?

~ 0 § ~ ~

ci

~ lQ ~ ci

h h i/'1.... H · 71 Î

100 150 200 250 300 350 100 150 200 250 300 350

SLICE10 SLICE15 ~ ~ 0

~ :!l

~ ~

!; ~ !; !Q ~ ..., 0

d ~ ó ~ ~

ci

~ ~ ~

"' "' 0 lil 0 lil :;l

0 0 o o I ~~ " ~ ol ~=;=-...,_. ....J 100 150 200 250 300 350

~ lil ~ lil

~ g ~ g "' 8 0 V "' 8 0 V

lQ 0 . 0

0 N f'l1'h_ lQ 0 . 0

0 N

100 150 200 250 300 350

~ : ~~~~~~ ~ : 100 150 200 250 300 350

Time (d)

I I

100 150 200 250 300 350

~ ~ ~ Cl ::; E ._ ci c

0 !Q ~ ci c

~V :-=-'f ' )- f- 0 ~ 0

100 150 200 250 300 350

(\ ~

lik... -J·

100 150 200 250 300 350

>-~ ..c

i!? 0

~

ci

lQ ci

a..

-- Brachionus -- Chaoborus -- Daphnia - Phytoplankton


95

Fig

ure 6

.5: F

ood

web

dyn

amic

s with

out c

hem

ical

expo

sure

as p

redi

cted

by t

he C

him

ERA

mod

el fo

r thr

ee en

viro

nmen

tal s

cena

rios (

scen

ario

s I,

II a

nd II

I) a

nd fo

ur se

lect

ed lo

catio

ns in

the

two-

pond

syst

em: a

t the

inflo

w o

f the

firs

t pon

d (S

lice

1), i

n th

e m

iddl

e of

the

first

pon

d (S

lice

5), b

etw

een

the

two

pond

s (Sl

ice

10) a

nd in

the

mid

dle

of th

e se

cond

pon

d (S

lice

15).

Onl

y th

e de

tritu

s-ba

sed

part

of t

he fo

od w

eb is

show

n.

Line

s and

shad

ed a

reas

indi

cate

the

aver

age

and

min

imum

and

max

imum

den

sitie

s, re

spec

tivel

y, o

f fiv

e ite

ratio

ns p

er sc

enar

io

( ~_l MM 5w) snl!Jlaa SL'O s·o sz·o I SL'O S'O sz·o 0 SL'O S'O SZ'O

~ ~ ~ I{) ~ ~ ~ ..--w ~ ~ ~ ü ::i ~ ~ ~ (/)

~ ~ ~

§ § §

oos oo• ooc ooz OOI 0 ooa 0001 oos 0001 009 ooz 0

SL'O s·o sz·o SL'O s·o sz·o SL'O s·o sz·o

~ ~ ~

0 ~ ~ ~ ..--"' w ~ ~ ~ a

ü -~

_J ~ ~ ~ D

(/) ~ ~ ~ -§ ê § ~ "' 2

oos oo• ooc ooz OOI 0 OOSI 0001 oos 0001 009 ooz 0 Q) "' E E SL'O s·o sz·o SL'O s·o sz·o SL'O s·o sz·o E

F "' (!)

~ ~ ~

I{) ~ ~ ~ w ~ ~ ~ "' ü :;,

== ::i ~ ~ ~ Q)

"' (/) <(

~ ~ ~

§ ê ê oos oo• ooc ooz OOI 0 OOSI 0001 oos 000~ 009 ooz 0

SL'O s·o sz·o SL'O s·o sz·o SL'O s·o sz·o

~ ~ ~

..-- ~ ~ ~ w ~ ~ ~ ü ::i ~ ~ ~ (/)

~ ~ ~

§ ê § oos oo• ooc ooz OOI 0 OOSI 0001 oos oom 009 ooz 0

a~uepunqv

Chapter 6

96

6.3.2.2. Carbendazim exposure

Direct effects

As a result of the low carbendazim concentrations, only minimal effects of carbendazim on the food

web dynamics were observed in scenario I (Figure 6.6-6.7). In scenario II and III, the expected negative

effects of carbendazim exposure to Daphnia and Gammarus densities were observed. High spatial

differences between the direct effects of carbendazim on Daphnia densities were observed in scenario

II and III: carbendazim effects were mainly observed in the first pond and were lower and occurred later

in the year in the second pond (slice 15), reflecting the lower carbendazim concentrations downstream.

As a result of the higher carbendazim concentrations, effects were more pronounced in scenario III than

in scenario II: after the first carbendazim application, the Daphnia densities decreased by 56% in the

first pond (slice 5) in scenario II while, in scenario III, Daphnia went extinct in the first pond (Figure

6.6). Recovery was only observed in the first pond in scenario II, where Daphnia densities increased

again after the first exposure peak (around day 250). The reached densities after recovery were, however,

only about 40% of the densities in the control scenario and the population peak occurred later (day 250

instead of day 200).

Carbendazim effects on Gammarus densities were observed in the whole system for both scenario II

and III (Figure 6.7): the strong population growth phase at the end of the year, observed in the control

scenarios (Figure 6.5), was no longer observed. Similar to the carbendazim effects on Daphnia, the

effects of carbendazim were lower and occurred later in the second pond, especially in scenario III.

Indirect effects

Because direct effects of carbendazim were nearly absent in scenario I, indirect effects were mainly

observed in scenario II and III on Asellus, Brachionus and Chaoborus. Indirect effects of carbendazim

were especially visible for the detritus feeders (Figure 6.7): the tolerant Asellus was able to become the

dominant species throughout in the system in both scenarios e.g. an increase in density by 329% in

scenario II. Reflecting the direct effects of carbendazim on Gammarus, the indirect effects occurred later

and the Asellus densities were lower in the second pond than in the first pond.

For the grazers, in scenario II, the densities of the more tolerant Brachionus were higher after

carbendazim exposure, although no shift in dominance occurred (Figure 6.6). As was the case for the

direct effects, spatial differences were observed: Brachionus initially increased in density in the first

pond after the first carbendazim application (day 160) but decreased again once Daphnia densities

recovered. In the second pond, carbendazim effects were nearly absent after the first carbendazim

application but an increase in Brachionus density was observed after the second application (day 250).

In scenario III, these spatial differences were even more pronounced. In the first pond, the carbendazim


97

application further increased the dominance of Brachionus leading to an increase in the number of peaks

in Brachionus density (from 5 to 7). In the second pond, however, indirect effects were absent.

Because Chaoborus can feed on both grazers, it was hard to formulate expectations for the indirect

effects of carbendazim. Apparently, the predator Chaoborus was affected indirectly in scenarios II and

III through the effects of carbendazim on the prey species Daphnia (Figure 6.6). As was the case for

Daphnia, effects were most pronounced in the first pond. In scenario II, Chaoborus densities decreased

by 32% in slice 5 but, following the recovery of Daphnia, stayed high for a longer period. In the second

pond, indirect effects were only observed at the end of the year by the absence of a second population

peak. In scenario III, Chaoborus followed the extinction of Daphnia in the first pond but no indirect

effects were observed in the second pond.

Chapter 6

98

Fig

ure

6.6:

Foo

d w

eb d

ynam

ics w

ith ca

rben

dazi

m e

xpos

ure a

s pre

dict

ed b

y the

Chi

mER

A m

odel

for t

hree

env

iron

men

tal s

cena

rios

(sce

nari

os

I, II

and

III)

and

four

sele

cted

loca

tions

in th

e tw

o-po

nd s

yste

m: a

t the

inflo

w o

f the

firs

t pon

d (S

lice

1), i

n th

e m

iddl

e of

the

first

pon

d (S

lice

5), b

etwe

en th

e tw

o po

nds

(Slic

e 10

) and

in th

e m

iddl

e of

the

seco

nd p

ond

(Slic

e 15

). O

nly

the

phyt

opla

nkto

n-ba

sed

part

of th

e fo

od w

eb is

sh

own.

Lin

es a

nd sh

aded

are

as in

dica

te th

e av

erag

e an

d m

inim

um a

nd m

axim

um d

ensi

ties,

resp

ectiv

ely,

of f

ive

itera

tions

per

scen

ario

.

( ~-1 MM 6W) UOptUeidOlÁ4d SL'O S'O SZ'O 0 I SL'O S'O Sl'O 0 I SL'O S'O SZ'O

~ ~ I{) ~ ..-- ~ ~

w ~ ü ::i ~ (/)

~

~

c

~

~

~

§ ~ ê c: osc osz OSI os 0 009 009 oo• ooz OOS I 0001 005 .9

"' c: SL'O s·o SZ'O SL'O s·o Sl'O SL'O s·o sz·o "' ä.

~

0 ~ ..--

~

~ j 0

~ >. .t::. a.

~

w ~ ü __J ~ (/)

~

§

~

~

~ ~

ê

~

~ .!,!!

~ c: .c: - g. § ~ 0

osc osz OSI os 0 009 009 oo• ooz OOSI 0001 005 Q)

SL'O s·o SZ'O SL'O s·o sc:·o SL'O s·o sz·o E F

~

I{) ~ w ~ ü

~

~

~

~ ., ;::

~ 2 0

"' ~ .c: ü

::i ~ (/)

~ ~

~ ~ ~

§ osc osz OSI os 0 008 009 oo• ooz

ê OOSI 0001 005

ê ~ c: .Q '5

SL'O s·o SZ'O SL'O s·o sc:·o SL'O s·o sz·o ~ IXl

~ ~ ~

..-- ~ ~ ~ w ~ ü ~ ~

::i ~ (/) ~ ~

~ ~ ~

§ ê § osc osz OSI os 0 008 009 oo• ooz OOSI 0001 005

a~uepunqv


99

Figure 6.7: Food w

eb dynamics w

ith carbendazim exposure as predicted by the C

himERA m

odel for three environmental scenarios (scenarios

I, II and III) and four selected locati ons in the two-pond system

: at the inflow of the first pond (Slice 1), in the m

iddle of the first pond (Slice 5), betw

een the two ponds (Slice 10) and in the m

iddle of the second pond (Slice 15). Only the detritus -based part of the food web is show

n. Lines and shaded areas indicate the average and m

inimum

and maxim

um densities, respectively, of five iterations per scenario.

Abundance

- - -0 200 600 1000 0 500 1500 2500 0 100 200 300 400 500

8 8 8

~ ~ ~

8 8 8

~ (/) r

~ ~ ~ 0 m

8 8 8 ......

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

0 200 600 1000 0 500 1500 2500 0 100 200 300 400 500

8 8 8

~ ~ ~ ):.

8 8 8

h (/)

"' r tb ê'

~ ~ ~ 0

"' m 8 8 8 CJ1

~ ~ ~ G) -i Q)

~ï 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

3 3 ct> 0 200 600 1000 0 500 1500 2500 0 100 200 300 400 500 Q)

2 "' 0: 8 8 8 ._..

~ ~

~ j~ (/)

0 8 8 r ~ "'

() Ë ~ ~ m (/1 ......

8 8 0

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

0 200 600 1000 0 500 1500 2500 0 100 200 300 400 500

8 8 8

~ ~

~ j~ (/)

8 8 r 0

~ ~ m ......

8 8 CJ1

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

Detritus (mg ww L-1)

Chapter 6

100

6.3.2.3. Chlorpyrifos exposure

Direct effects

As expected, direct effects of chlorpyrifos were observed for Chaoborus, Daphnia and Gammarus

(Figures 6.8 and 6.9). The magnitude of the effects was directly linked to the exposure concentrations

that were predicted by the ChimERAfate model (Figure 6.3): clear differences in effects could be

observed between environmental scenarios (I, II and III) and between different locations in the system.

Effects in scenario I were limited for the phytoplankton-based part of the food web (Figure 6.8). Because

of the higher and prolonged exposure to chlorpyrifos, effects were more pronounced in scenario II and

III than in scenario I. The fast decrease in chlorpyrifos concentrations away from the emission point

resulted in effects mainly occurring in the inflow stream (slice 1) and the first pond (slice 5). In scenario

II, the chlorpyrifos effects on Daphnia were clear in the first pond but recovery occurred quickly: the

population peak occurred later (day 250 instead of day 200) and was lower (551 ∙ 102 ind L-1 instead of

733 ∙ 102 ind L-1). A second population growth period at the end of the year (Figure 6.4) was no longer

observed. For scenario II, chlorpyrifos effects on the predator Chaoborus were also mainly observed in

the first pond e.g. the maximum density in slice 5 dropped from 397 ind L-1 to 114 ind L-1. In scenario

III, the direct effects of chlorpyrifos were lower and the same spatial trend was observed: clear effects

in the first pond and no effects in the second pond.

For the detritus feeders, the densities of Gammarus were negatively affected in scenario I and effects

were observed throughout the system (Figure 6.9). Gammarus completely disappeared in the first pond

and the observed population growth in the second pond was only a fraction (<10%) of the control

scenario. In scenario II, chlorpyrifos exposure led to the elimination of Gammarus in the first pond and

severely affected the densities in the second pond, where maximum densities dropped from 1578 ind L-

1 to 1066 ind L-1 for slice 15. In scenario III, although chlorpyrifos effects on Gammarus were lower

than in scenario II, the differences between the first pond and the second pond were even greater than

in scenario II: while chlorpyrifos effects were still observed in the first pond (52% reduction in density),

these were negligible in the second pond.

Indirect effects

The expected indirect effects of chlorpyrifos were observed and reflected the spatial pattern of the direct

effects (Figure 6.8-6.9). Indirect effects on Brachionus were absent in scenario I (Figure 6.8). In scenario

II, the indirect effects on Brachionus only occurred in the inflow stream and the first pond and were

limited because of the recovery of Daphnia after chlorpyrifos exposure. Chlorpyrifos effects on Daphnia

also influenced the phytoplankton dynamics in scenario II: the first growth season of phytoplankton

lasted longer in the first pond (from day 150 to day 230 versus from day 150 to day 190). In scenario


101

III, indirect effects on Brachionus were again limited to the inflow stream and the first pond and only

resulted in a small increase in Brachionus density (<10%).

In scenario I, the densities of Gammarus were negatively affected but indirect effects on its competitor

Asellus were not observed (Figure 6.9). In scenario II, however, Asellus densities increased strongly

after chlorpyrifos emission in the inflow stream, first pond and connecting stream, leading to maximum

densities of 5708 ind L-1 in slice 5 versus 116 ind L-1 in the control scenario. The sharp population

decline in Asellus observed in the first pond around day 250 was a result of starvation. In scenario III,

the spatial pattern of the direct effects on Gammarus was confirmed and (low) indirect effects of

chlorpyrifos on Asellus were observed in slices 1 and 5 only.

Chapter 6

102

Fig

ure

6.8:

Foo

d w

eb d

ynam

ics w

ith c

hlor

pyri

fos e

xpos

ure

as p

redi

cted

by

the

Chi

mER

A m

odel

for t

hree

env

iron

men

tal s

cena

rios

(sce

nari

os

I, II

and

III)

and

four

sele

cted

loca

tions

in th

e tw

o-po

nd s

yste

m: a

t the

inflo

w o

f the

firs

t pon

d (S

lice

1), i

n th

e m

iddl

e of

the

first

pon

d (S

lice

5), b

etwe

en th

e tw

o po

nds

(Slic

e 10

) and

in th

e m

iddl

e of

the

seco

nd p

ond

(Slic

e 15

). O

nly

the

phyt

opla

nkto

n-ba

sed

part

of th

e fo

od w

eb is

sh

own.

Lin

es a

nd sh

aded

are

as in

dica

te th

e av

erag

e an

d m

inim

um a

nd m

axim

um d

ensi

ties,

resp

ectiv

ely,

of f

ive

itera

tions

per

scen

ario

.

( ~-1 MM 6W) UOptUeidOlÁ4d SL'O S'O SZ'O 0 I SL'O S'O SZ'O 0 I SL'O S'O SZ'O

~ I{) ~ ..-- ~

~

~

w ~ ü ::i ~ (/)

~

~

c J

~

~

~

§ ~ ê c: ost osz OSI os 0 009 009 00> ooz OOSI 0001 oos .9

"' c: SL'O s·o sz·o SL'O s·o sz·o SL'O s·o sz·o "' ä.

~ ~ 0

~ >. .t::. a.

0 ~ ..--w ~ ü __J ~

~ ~

~

~

~

~

~ (/)

~

§

~

ê

.!,!!

~ c: .c: - g. § ~ 0

ost osz OSI os 0 009 009 00> ooz OOSI 0001 oos Q)

SL'O s·o sz·o SL'O s·o sz·o SL'O s·o sz·o E F

~

I{) ~ w ~ ü

~

~ 2: ~

~ ., ;::

~ 2 0

"' ~ .c: ü

::i ~ (/)

~

~ .:- .f..

~

~

~

§ ost osz OSI os 0 009 009 001> ooz

ê l OOSI 0001 oos

ê ~ c: .Q '5

SL'O s·o sz·o SL'O s·o sz·o SL'O s·o sz·o ~ IXl

~ ~ ~

..-- ~ ~ ~ w ~ ü ~ ~

::i ~ (/) ~ ~

~ ~ ~

§ ê § ost osz OSI os 0 009 009 001> ooz OOSI 0001 oos

a~uepunqv


103

Figure 6.9: Food w

eb dynamics w

ith chlorpyrifos exposure as predicted by the Chim

ERA model for three environm

ental scenarios (scenarios I, II and III) and four selected locations in the tw

o -pond system: at the inflow

of the first pond (Slice 1), in the middle of the first pond (Slice

5), between the tw

o ponds (Slice 10) and in the middle of t he second pond (Slice 15). O

nly the detritus-based part of the food web is shown.

Lines and shaded areas indicate the average and minim

um and m

aximum

densities, respectively, o f five iterations per scenario.

Abundance

- - -0 200 400 600 800 0 2000 4000 6000 0 50 100 150 200

8 8 8

~ ~ ~

;jF (/) 8 8 r

~ ~ 0 m

8 8 ...... 8

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

0 200 400 600 800 0 2000 4000 6000 0 50 100 150 200

8 8 8

~ ~ ~ ):.

8 8 :jp (/)

"' r tb

d ê' ~ ~

0 "' m

8 8 1 CJ1

~ ~ ~ G) -i Q)

~ï 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 3

3 ct> 0 200 400 600 800 0 2000 4000 6000 0 50 100 150 200 Q)

2 "' 0: 8 8 8 ._..

~ ~

~ j~ (/)

0 8 8

~ r

~ "' ()

Ë ~ ~ m (/1 ......

8 8 0

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

0 200 400 600 800 0 2000 4000 6000 0 50 100 150 200

8 8 8

~ ~ ~ (/)

8 8 8 r 0

~ ~ ~ m ......

8 8 8 CJ1

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75


Chapter 6

104

6.3.2.4. Pyrene exposure

Direct effects

In scenario I, Daphnia population dynamics exposed to pyrene (Figure 6.10) followed the same pattern

as in the control scenario (Figure 6.4). However, the maximum observed densities of Daphnia were

lower than those in the control scenario (131 ∙ 102 ind L-1 versus 205 ∙ 102 ind L-1). In scenario II and

especially scenario III, effects on Daphnia were higher than in scenario I and eventually daphnids were

eliminated from the food web. Following the exposure pattern, effects occurred first close to the

emission point (slice 1). Direct effects on Gammarus were absent (Figure 6.11): in none of the scenarios

was there a decrease in Gammarus densities that could be directly related to the pyrene concentration

and Gammarus concentrations were even higher in scenario III. The observed effects on Gammarus

were mainly indirect effects (see below).

Indirect effects

In scenario I, although expected, the decreased Daphnia densities did not lead to higher Brachionus

densities (Figure 6.10). In scenario II, the expected indirect effects of pyrene were observed: Brachionus

was clearly more abundant than in the control scenario and three peaks in density were observed

throughout the year. In scenario III, after the early disappearance of Daphnia, Brachionus was even

more dominant than in the control scenario: peaks in population density were even higher and the

number of peaks in density increased from 5 to 8. The predator Chaoborus was affected indirectly by

pyrene in all three scenarios through the lower prey density. In scenario I, Chaoborus densities decreased

from 314 ind L-1 in the control scenario to 158 ind L-1. Both in scenario II and III, despite the increase

in the other prey species Brachionus, Chaoborus disappeared once Daphnia was eliminated from the

food web by pyrene.

The combination of direct effects on Daphnia and indirect effects on Brachionus altered the

phytoplankton dynamics (Figure 6.10). This, in turn, affected the detritus dynamics, resulting in

unexpected indirect effects on the detritus feeders (Figure 6.11). In scenario I, the altered detritus

dynamics resulted in the absence of Gammarus population growth at the end of the year. In scenario II,

the phytoplankton concentration was more constant in the ponds between day 175 and 275, resulting in

more constant detritus concentrations and the absence of a peak in detritus concentrations compared to

the control scenario (Figure 6.5). This resulted in lower Gammarus densities (only 464 ind L-1 compared

to 1579 ind L-1 in the control scenario) but did not affect Asellus densities. In scenario III, the number

of phytoplankton growth periods increased from 6 to 9, which was reflected in the detritus dynamics.

As a result, both Gammarus and Asellus densities increased in comparison to the control scenario, from

1034 ind L-1 and 405 ind L-1 to 1401 ind L-1 and 545 ind L-1, respectively.


105

Fig

ure

6.10

: Foo

d w

eb d

ynam

ics

with

pyr

ene

expo

sure

as

pred

icte

d by

the

Chi

mER

A m

odel

for

thre

e en

viro

nmen

tal s

cena

rios

(sce

nari

os I,

II

and

III)

and

four

sele

cted

loca

tions

in th

e tw

o-po

nd sy

stem

: at t

he in

flow

of t

he fi

rst p

ond

(Slic

e 1)

, in

the

mid

dle

of th

e fir

st p

ond

(Slic

e 5)

, be

twee

n th

e tw

o po

nds (

Slic

e 10

) and

in th

e m

iddl

e of

the

seco

nd p

ond

(Slic

e 15

). O

nly

the

phyt

opla

nkto

n-ba

sed

part

of t

he fo

od w

eb is

show

n.

Line

s and

shad

ed a

reas

indi

cate

the

aver

age

and

min

imum

and

max

imum

den

sitie

s, re

spec

tivel

y, o

f fiv

e ite

ratio

ns p

er sc

enar

io.

( ~-1 MM 6W) UOptUeidOlÁ4d Sl'O s·o SZ'O 0 I Sl'O S'O sz·o 0 I Sl'O S'O SZ'O

~ ~ ~ I{) ~ ~ ~ ..--w ~ ~ ~ ü ::i ~ ~ ~ (/)

~ ~ ~

§ § § c:

OSI OOI os ooz OOI OS 0 OOSI 0001 oos .9 "' c:

Sl'O s·o SZ'O Sl'O s·o sz·o Sl'O s·o sz·o "' ä. 0

~ ~ ~ >. .t::. a.

0 ~ ~ ~ ..--w ~ ~ ~ ü __J ~ ~ ~ (/) .!,!!

~ ~ ~ c: .c: - g. § ê § ~ 0

OSI OOI os ooz OOI OS 0 OOSI 0001 oos Q)

Sl'O s·o SZ'O Sl 'O s·o sz·o Sl'O s·o sz·o E F

~ ~ ~ ., ;::

~ ~ ~ 2 I{) 0

"' w ~ ~ ~ .c: ü

ü ::i ~ ~ ~ (/)

~ ~ ~

§ ê ê ~ c: .Q

OSI OOI os ooz 00~ os 0 OOSI 0001 oos '5 Sl'O s·o SZ'O Sl'O s·o sz·o Sl'O s·o sz·o ~

IXl

~ ~ ~

..-- ~ ~ ~ w

~ ~ ~ ~ ü

::i ~ ~ ~ (/)

~ ~ ~

§ ê § OSI OOI os ooz OOI os 0 OOSI 0001 oos

a~uepunqv

Chapter 6

106

Figure 6.11: Food w

eb dynamics w

ith pyrene exposure as predicted by the Chim

ERA model for three environm

ental scenarios (scenarios I, II and III) and four selected locations in the tw

o -pond system: at the inflow

of the first pond (Slice 1), in the middle of the first pond (Slice 5),

between the tw

o ponds (Slice 10) and in the middle of the second pond (Slice 15). O

nly the detritus -based part of the food web is show

n. Lines and shaded areas indicate the average and m

inimum

and maxim


Abundance

- - -500 1000 1500 0 100 300 500 0 50 100 150 200

8 8 8

~ ~ ~

8 8 ; j~ (/) r

~ ~ 0 m

8 8 ...... 8

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

500 1000 1500 0 100 300 500 0 50 100 150 200

8 8 8

~ ~ ~ ):.

8 8 :JP (/)

"' r tb ê'

~ ~ 0

"' m 8 8 CJ1

~ ~ ~ G) -i Q)

~ï 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 3

3 ct> 0 500 1000 1500 0 100 300 500 0 50 100 150 200 Q)

2 "' 0: 8 8 8 ._..

~ ~

~ j~ . (/)

0 8 8 r ~ "'

() Ë ~ ~ m (/1 ......

8 8 0

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75

500 1000 1500 0 100 300 500 0 50 100 150 200

8 8 8

~ ~

~ jp (/)

8 8 r 0

~ ~ m ......

8 8 CJ1

~ ~ ~

0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75



107

6.3.2.5. Exposure to the mixture of all three chemicals

Direct effects

In scenario I, the pattern in Daphnia densities was similar to the control scenario (Figure 6.4) but the

maximum densities were lower (Figure 6.12). In scenario II, Daphnia was eliminated quickly in the

whole system with effects appearing earlier close to the emission source (slice 1). Daphnia disappeared

from the first pond by day 200 and from the second pond by day 250. In scenario III, the effects of the

mixture occurred even faster, with Daphnia disappearing from the first pond by day 150.

For Gammarus in scenario I, direct effects of the mixture on the densities were limited and only occurred

after day 250 (Figure 6.13). In scenario II, the effects were more clear, although they only started

occurring after carbendazim and chlorpyrifos were first applied (day 180): Gammarus disappeared from

the system and this occurred first close to the emission point. Effects of the mixture on Gammarus were

also clear in scenario III but Gammarus was able to persist longer in the second pond compared to

scenario II.

Direct effects of the mixture for Chaoborus are expected to be similar to the direct effects of chlorpyrifos

as this species is tolerant for the other two chemicals. Direct effects of the mixture on Chaoborus

densities were not observed in scenario I (Figure 6.12). In scenario II, a sharp decrease in Chaoborus

density is observed in the first pond (slice 5) after the application of chlorpyrifos and carbendazim (day

180). Direct effects in the second pond were limited. In scenario III, direct effects of the mixture were

no longer observed and the population dynamics of Chaoborus were determined by the indirect effects

of the mixture (see below).

Indirect effects

For Brachionus in scenario I, the indirect effects were negligible (Figure 6.12) because the densities

were very similar to the control scenario (Figure 6.4). In scenario II, following the disappearance of

Daphnia after chemical exposure, densities of the tolerant Brachionus increased in all slices and showed

three distinct population peaks. In scenario III, indirect effects of exposure to the mixture of three

chemicals resulted in an increase in Brachionus densities. Also, the frequency of the peaks in Brachionus

density increased from 5 to 8 compared to the control scenario.

Through the strong direct and indirect effects of the mixture on the grazers, phytoplankton dynamics

were altered. Because of the close link between phytoplankton and detritus, this resulted in indirect

effects of the mixture on the detritus feeders. In scenario I, the altered detritus concentrations resulted

in the absence of Gammarus growth at the end of the year (Figures 6.5 and 6.13). This did, however,

not result in an increase in Asellus densities. In scenario II, the exposure to the mixture resulted indirectly

in increased Asellus densities in the whole system and this increase was highest in the first pond (slice

Chapter 6

108

5). In the case of scenario III, the increase of Asellus was only observed for the first pond. Also in

scenario III, because of the altered detritus dynamics, Gammarus was able to reach higher densities in

the second pond (1314 ind L-1) compared to the control scenario (1034 ind L-1).

In scenario I, Chaoborus densities were lower (Figure 6.12) than the control scenario (Figure 6.5)

because of the decreased densities of Daphnia. In scenario II, in addition to the direct effects of the

mixture, Chaoborus was indirectly affected through the effects on Daphnia and population densities

were never higher than 100 ind L-1. In scenario III, the effects of the mixture on Daphnia resulted in the

elimination of Chaoborus even before carbendazim and chlorpyrifos were applied (day 180).


109

Figure 6.12: Food w

eb dynamics exposed to three chem

icals as predicted by the ChimERA m

odel for three environmental scenarios (scenarios

I, II and III) and four selected locations in the two -pond system

: at the inflow of the first pond (Slice 1), in the m

iddle of the first pond (Slice 5), between the tw

o ponds (Slice 10) and in the middle of the second pond (Slice 15). O

nly the phytoplankton -based part of the food web is

shown. Lines and shaded areas indicate the average and m

inimum

and maxim


Abundance

- - -500 1000 1500 0 50 150 250 350 0 50 100 150

8 8 8

~ ~ ~

8 8 ;jW (/) r 0

~ ~ m 8 8 ......

8

~ ~ ~ I

tiJ iiJ

0 0.25 0.5 0.75 1 0 025 0.5 0.75 1 0 0.25 0.5 0.75

g. 0 500 1000 1500 0 50 150 250 350 0 50 100 150 o· => 8 8 8 ~

~ ~ ~

8 8

: J~ (/) r

~ ~ 0

() m ::r Q)

CJ1 0 8 8 c::r 0 2 ~ ~ ~ "'

-i ~ï

0 0.25 0.5 0.75 1 0 025 0.5 0.75 1 0 0.25 0.5 0.75

ct> 0 500 1000 1500 0 50 150 250 350 0 50 100 150

I:J 0: 8 8 8 ~ -::r ~ ~

~ j~ =>

(/) iiï

8 8 r ()

"' ~ ~ m ......

8 8 0 -o

~ ~ ~ :::r 'S. 0 "0 iiï 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 ::>

" ö 0 500 1000 1500 0 50 150 250 350 0 50 100 150

::> 8 8 8

~ ~

~ j~ (/)

8 8 r 0

~ ~ m ......

8 8 CJ1

~ ~ ~

0.25 0.5 0.75 1 0 025 0.5 0.75 1 0 0.25 0.5 0.75

Phytoplankton (mg ww L -1)

Chapter 6

110

Figure 6.13: Food web dynamics exposed to three chemicals as predicted by the ChimERA model for three environmental scenarios (scenarios I, II and III) and four selected locations in the two-pond system: at the inflow of the first pond (Slice 1), in the middle of the first pond (Slice 5), between the two ponds (Slice 10) and in the middle of the second pond (Slice 15). Only the detritus-based part of the food web is shown. Lines and shaded areas indicate the average and minimum and maximum densities, respectively, of five iterations per scenario.

SLICE 1 SLICE 5 SLICE10 SLICE15 ~ ~ ~ ~

~ ~ ~ ~

~ !; ~ !; ~ !; ~ ..., 0

~ ~ ~ ~

ó ó ó ó

ê 8 8 8 ~ ~

\ 1~ ~ ~ ~ ~ ~

"' "' "' :;l ~ 0 ~ 0 ~ 0 ~

0 0 0 0 0 0 i I I I I

100 150 200 250 300 350 100 150 200 250 :300 350 100 150 200 250 300 350 100 150 200 250 300 350 .-I

~ ...J

Q)

:1 K! ~

F1 f 2 I

fH1 f;! (.)

c ó

ro ~ 8 ~ 8 "0 11 ei "' ei "'

_{1 :;J E c -:::J liJ 8 liJ 8 liJ C/) .0 b. <( ,.1\ 1'.~

ei ~ ei ~

~ ei :::J -' ï ::::: 0 -

Q) 100 150 200 250 300 350 100 150 200 250 300 350 100 150 200 250 300 350 100 150 200 250 300 350 0

0

~ - ~ ~ ~ - ~ i!? i!? i!?

r , ó ó ó ó

:t~ 8 8 8

111 \t) ~ ~ :!! ~ :!! ~

ei ei ei ei

~ § ~ ~ ~ ~ liJ ei

0 0 0 0 0 0

100 150 200 250 300 350 100 150 200 250 300 350 100 150 200 250 300 350 100 150 200 250 300 350

Time (d)

- Aseflus - Gammarus - Delritus


111

6.4. Discussion

6.4.1. ChimERAfate simulations

When comparing the fate model predictions among the different scenarios, the residence time –reflecting

the removal of the chemical from the ponds via with the water (out)flow – was the most important factor.

Higher residence times resulted in less dilution of the chemical i.e. more chemical was emitted in the

same volume of water. This explains why higher water concentrations were observed (for all chemicals)

in scenario III (residence time of 100 days) compared to the concentrations in scenario II and I (residence

time of 25 and 5 days, respectively). As discussed in the expectations section (6.2.2.1), sediment

accumulation of the chemical also increased with water residence time, decreasing the concentration in

the water. Apparently, sediment accumulation was not strong enough to compensate for the positive

effect of retention time on the water concentration. However, the effects of sediment accumulation were

visible in the spatial patterns within the system: because most of the chemical accumulated in the

sediment, the water concentrations of the hydrophobic chemical chlorpyrifos were much lower in the

second pond than in the first pond and this was more expressed in scenarios with long residence times.

The expected negative effect of a higher temperature – and the resulting increased phytoplankton and

detritus biomass – and a higher organic matter concentration on the water concentration were not

observed. These two processes were apparently not strong enough to compensate for the positive effect

of a higher residence time.

The fate simulations for carbendazim differed substantially from the expectations (6.2.2.1). Despite the

short peaks in emission, concentrations remained high for much longer than expected further away from

the emission point in scenarios with higher residence times. Although no accumulation in sediment

occurred for this hydrophilic chemical, diffusion into the sediment pore water and later again into the

water column was an important fate process for carbendazim. This allowed for more constant dissolved

water concentrations. More time for equilibrium between the sediment pore water and the water column

was available in systems with high water residence times. This resulted in more diffusion into the

sediment, explaining the stronger decrease in carbendazim water concentration further downstream in

scenarios III than in scenario II and I.

As expected, because of the hydrophobic nature of chlorpyrifos, accumulation in the sediment occurred

and this led to constant water concentrations even when emission had stopped. This was best visible in

slice 1, where the emission occurred: once emission stopped, the water concentration quickly decreased

to a low but relatively constant ambient concentration. Because most of the chlorpyrifos accumulated in

the sediment, maximum chlorpyrifos concentrations were lower and reached later further downstream

in the system, especially in scenarios with high water residence time. Simulated pyrene concentrations

Chapter 6

112

became stable over time, as expected, because of the constant emission strength. Water residence time

and sediment accumulation determined how fast water concentrations stabilized: the emitted pyrene

flowed downstream slower with increased water residence time and accumulation in the sediment

further prolonged the stabilization of the water concentrations.

With the ChimERAfate model, we clearly showed that spatiotemporal differences in exposure were

determined by the physico-chemical properties of the chemical and the environmental variables

considered here (hydrodynamics, trophic state and temperature). Although environmental fate processes

are well studied and can be described accurately with equations, the predicted patterns in exposure

sometimes differed from our expectations e.g. for carbendazim. The relative importance of each

environmental process is not always clear a priori and can change with the environmental conditions.

Fate models are thus ideal tools to identify which chemicals will show a strong spatial pattern and in

which environmental conditions spatial patterns will be most prominent. Identifying whether chemicals

will only be present in high concentrations close to the emission source or also further downstream is

an invaluable tool to predict the risk of the chemical.

6.4.2. ChimERAfoodweb without chemical exposure

Both phytoplankton growth and maximum phytoplankton concentration increased with the increase in

trophic state and temperature. As a result, grazer densities increased rapidly in eutrophic, high

temperature environments such as scenario III. The higher trophic state and temperature in scenario III

compared to the other two scenarios also led to a more dynamic system, with a typical cyclic pattern of

phytoplankton growth, increase in grazer densities, phytoplankton decline and decrease in grazer

densities. These faster food web dynamics in mesotrophic systems compared to those in oligotrophic

systems were predicted before for a similar food web using ordinary differential equations to describe

the biomass of the species in the food web (De Laender et al. 2015). Because Brachionus has a much

shorter life cycle than Daphnia, the rotifers were able to increase in density more rapidly during

phytoplankton growth periods, leading to the observed shift in dominance in scenario III compared to

scenario I and II. However, this does not seem to reflect observations in real ecosystems, where daphnids

are known to outcompete rotifers (Gilbert 1985; MacIsaac and Gilbert 1989). Other competitive

mechanisms between large daphnids and rotifers have been suggested in literature e.g. through

mechanical interference (MacIsaac and Gilbert 1989). As shown in Chapter 4, accounting for

competitive exclusion was sufficient to simulate the outcome of competition between Daphnia magna

and Brachionus calyciflorus for the experimental conditions used in Chapter 2. However, alternative

mechanisms of competition could be more important when environmental conditions are more realistic

than the controlled conditions used in Chapter 2. Because the relative importance of such alternative

modes of competition was not known, they were therefore not implemented in the DEBkiss IBMs.

113

The current food web dynamics are clearly bottom-up regulated: grazer dynamics were limited by the

phytoplankton, detritus feeders followed the detritus pattern and prey dynamics. Although they are able

to prey on both Brachionus and Daphnia, Chaoborus was most abundant when Daphnia was dominant

i.e. in scenario II. Seemingly, Chaoborus was thus not able to sustain high population densities in

scenario III when maximum Brachionus densities were high. However, Brachionus densities also

fluctuated strongly in scenario III. The lower Chaoborus densities in scenario III were probably related

to the absence of a reserve compartment in the DEBkiss theory. When prey populations were high,

energy uptake by Chaoborus was also high and this energy was immediately spent. When prey densities

were low, the energy uptake by Chaoborus was insufficient. In reality, Chaoborus would first utilize

energy from the reserve buffer in that situation (Kooijman 2010). However, since a reserve compartment

is lacking in DEBkiss, energy is instead taken from the reproduction buffer or, when all energy in the

reproduction buffer is spent, from structural biomass. This led to a much faster response to food shortage

compared to reality, as discussed in Chapter 4. Daphnids show much less extreme population dynamics,

serving as better prey species for the Chaoborus DEBkiss IBM.

Even without chemical exposure, there were notable spatial differences in food web dynamics: except

for Brachionus, population densities were lower in the streams than in the ponds. This is a direct result

of the way movement was implemented. Movement was implemented as a species-specific chance to

move to a neighbouring water patch. Patches located in a pond were surrounded by more water patches

than patches in a stream. Patches in a pond thus had a higher chance of receiving new individuals than

patches in a stream. This results in highly different population dynamics between ponds and streams for

mobile species. For example, the probability for Daphnia 1 to move to a neighbour water patch was 1 d-

1. As a result, daphnid density was much higher (e.g. up to five times for scenario II) in ponds than in

streams. The lower Daphnia and Chaoborus densities in the streams meant less competition and

predation pressure, allowing Brachionus, which had a lower movement chance (0.1 d-1), to reach higher

densities in the streams than in the ponds e.g. up to two times higher for scenario II.

6.4.3. ChimERAfoodweb with chemical exposure

The magnitude of effects was directly linked to the concentrations that were predicted by the

ChimERAfate model: clear differences in effects could be observed between the different environmental

scenarios (I, II and III) and between the different locations in the system. Recovery of the phytoplankton-

based part of the food web was observed for chlorpyrifos and carbendazim after the first emission peak

(around day 200). The second emission peak, combined with the decreased food availability and the

increased competition with the tolerant species (Brachionus), prevented recovery during the simulated

period. Unsurprisingly, recovery was not observed for pyrene exposure because the concentration was

relatively constant throughout the year. Recovery of the detritus-based part of the food web was

observed to be much more difficult. This was mainly due to the longer reproduction times of the detritus

Chapter 6

114

feeders: i.e. these were at least 40 and 80 days between reproduction events for Gammarus and Asellus,

respectively. Slower recovery of species with long life cycles or generation times has been raised as a

concern for ERA before (De Lange et al. 2010; Rubach et al. 2010). By including species with different

life cycle times or by adjusting the currently implemented generation times, the ChimERA model can

be a tool to assess how life cycle duration affects food web recovery.

Simulations performed using the ChimERA model showed a strong spatial pattern, with effects of

chemicals being absent, smaller or delayed in the second pond compared to those observed in the first

pond. A good example of the absence of effects in the second pond is the exposure to chlorpyrifos in

scenario II: the water concentrations in the first pond clearly caused a delay in daphnid population

growth while the food web dynamics in the second pond were completely unaffected. Smaller effects in

the second pond than in the first pond were, for example, observed for carbendazim in scenario III:

Gammarus densities were less affected in the second pond compared to the first pond. Delayed effects

were clearly visible for pyrene in scenario III: Daphnia populations in the second pond also became

extinct but this occurred much later than in the first pond. Also, because Daphnia and Chaoborus tended

to remain in the ponds and were much less abundant in the streams, the food web dynamics in the streams

were already more determined by the tolerant Brachionus and effects of the chemical were generally

less severe there. This highlights the need in a spatial landscape to account for the local conditions.

Environmental variables such as temperature, light conditions and water flow strength are distributed

heterogeneously in a landscape. Environmental conditions determine which species are successful and

local food webs can thus differ greatly in their structure and composition. Chemical risk for a landscape

is not homogeneously distributed but local differences will exist because exposure and food web

vulnerability differ locally. The ChimERA model, although only applied here for a two-pond system, is

a good example of how an integrated fate and effect model can be used to differentiate the local risks

within a spatial system.

Indirect effects were prevalent in the ChimERA simulations and included indirect effects on

competitors, on predators and on a food source (i.e. phytoplankton and detritus). Chemical effects on

the sensitive species often led to increased densities of the tolerant species which became dominant in

some cases. Especially the indirect effect of chlorpyrifos on Asellus densities in scenario II was striking.

Brachionus typically also increased in density after chemical effects on Daphnia had occurred. This

effect was, however, less drastic than those observed on Asellus and mainly occurred in scenario II. In

scenario I, rotifers were quickly outcompeted by the daphnids, even before chemical effects became

apparent, preventing possible indirect effects on Brachionus. In scenario III, rotifers were already quite

high in density and indirect effects only resulted in small increases in density. A shift in dominance

between competitors is often observed after chemical exposure. For example, increases in rotifer

abundance after chemical effects on cladocerans were observed after the application of fenvalerate in

lake enclosures (Day et al. 1987). However, rotifers can also be outcompeted by cladocerans after

115

chemical exposure e.g. the cladoceran Moina macrocopa outcompeted the sensitive rotifer Brachionus

calyciflorus after exposure to cadmium (Gama-Flores et al. 2006). Other examples of shifts in

dominance after chemical exposure include the replacement of green algae by cyanobacteria and

diatoms after fomesafen application (Caquet et al. 2005) or the increase in tolerant rotifers species after

carbendazim effects on cladocerans and copepods (Van den Brink et al. 2000). Theoretically, it makes

sense that an inferior competitor can be successful when the superior competitor is affected by chemical

stress because less resources are consumed by the superior competitor. However, this is not always

observed in experiments (Chapter 2). In the ChimERA simulations, even when the superior competitor

was affected, other conditions were also important. For example, in the simulations performed for

pyrene exposure in scenario I, effects of pyrene on Daphnia densities were not sufficiently high to

prevent the elimination of Brachionus. Also, the food concentrations in scenario I were too low to sustain

high densities of Brachionus, which was demonstrated in the streams where competition with Daphnia

was almost absent but where Brachionus densities were still close to zero. Models like the ChimERA

model can help to understand and predict when and why indirect effects occur.

Indirect effects of carbendazim and pyrene were observed on the predator Chaoborus. Although the

maximum densities of Brachionus increased after chemical exposure, these rotifer densities fluctuated

much more than Daphnia densities. With the current Chaoborus DEBkiss IBM implementation, it was

difficult to simulate stable population densities when food availability is highly dynamic, as discussed

before (see 6.4.2). Because of the high fluctuations in Brachionus densities, Brachionus was thus a less

suitable prey for Chaoborus, leading to the observed indirect effect on Chaoborus densities: reduced

population densities when Daphnia density was negatively affected by chemical stress and even

extinction e.g in the case of pyrene exposure in scenario III. Altered Daphnia densities due to chemical

stress lowered the Chaoborus densities but could also lead to a longer period of stable predator densities

when daphnid densities recovered from chemical exposure e.g. after carbendazim exposure in scenario

II.

Phytoplankton is the only source of detritus in the fate model. Therefore, it is expected that a chemical

exposure affecting grazer dynamics will lead to changes in phytoplankton dynamics and as a result, in

different detritus dynamics. For example, due to the pyrene exposure in scenario II, the phytoplankton

dynamics shifted from two distinct peaks at day 175 and 250 in the ponds to a more constant

phytoplankton concentration with four smaller peaks between day 150 and 250. This resulted in different

detritus dynamics, ultimately also affecting the detritus feeders, in this case by preventing the strong

increase in Gammarus density at the end of the year. Interestingly, shifts in grazer densities did not

necessarily affect the detritus feeders negatively. For pyrene in scenario III for example, the dominance

shift to Brachionus increased the amount of detritus available, resulting in higher Asellus and Gammarus

densities. In reality, however, the main food sources of Asellus and Gammarus are not from

phytoplankton origin. The diet of Asellus and Gammarus is much more diverse with leaf litter from

Chapter 6

116

external sources (terrestrial plants) often cited as the most important food source (Moore 1975). This

was not considered in the ChimERA model and Asellus and Gammarus were thus much more dependent

on phytoplankton dynamics than in reality. However, this did allow us to study the interactions between

phytoplankton, detritus and associated consumers.

For the grazers and predator in the food web, the overall effect(s) of the mixture seemed to be dominated

by the effects of pyrene: effects of the mixture on the grazers corresponded mostly with the effects of

pyrene alone, with effects occurring almost immediately after the start of the simulated exposure. For

the detritus feeders, however, this was not the case: direct effects of pyrene on Gammarus were limited

in the pyrene only exposure, but effects of the mixture were clearly visible leading to a dominance of

Asellus. Identification of the dominant chemical in the mixture was done by comparing which effects of

the single exposure corresponded most to the effects of the mixture. For more complex food web models

or mixtures, where more effects can occur, this approach is insufficient and alternative approaches need

to be considered e.g. similarity indices such as the Bray-Curtis index (Bray and Curtis 1957). Similarity

indices describe the composition of the community/food web with a single number. For example, the

Bray-Curtis indices of the communities exposed to the mixture and of those exposed to the chemicals

in the mixture individually could be compared. The Bray-Curtis index of the individual chemical closest

to the index of the mixture then indicates in which community the effects of the individual chemical are

most similar to the effects of the mixture i.e. which effect dominates.

Mixture toxicity at the community level can be very different from mixture toxicity at the level of

individuals and populations. For example, the effects of a herbicide and an insecticide, affecting

phytoplankton and grazers, respectively, are not necessarily additive (i.e. the sum of the individual

effects) at the food web level. Synergistic effects at the food web level (i.e. higher effects than what is

expected from the individual effects) could occur when the grazers are affected both by direct effects of

the insecticide and by indirect effects due to herbicide effects on the phytoplankton. Antagonistic effects

(i.e. smaller effects than what is expected from the individual effects) could occur when the

phytoplankton is affected by the herbicide but is less grazed upon because of the adverse insecticide

effects on the grazers. Models like the ChimERA model offer a unique way to account for this.

6.4.4. The ChimERA model as a risk assessment tool

The ChimERA model demonstrates how integrating a fate and food web model offers insights that

cannot be obtained when running both models separately. Experimental/field data to compare our model

simulations with were not available. However, the underlying models – ChimERAfate and

ChimERAfoodweb – have been tested: the ChimERAfate model was successfully applied in three case

studies for a pond system (Morselli et al. 2015) and the application of the DEBkiss IBMs for competing

species – used in the ChimERAfoodweb model –was evaluated in Chapter 4. The simulations of the

117

ChimERA model were therefore considered realistic although it cannot be excluded that certain

processes or mechanisms important for the integrated model were not incorporated.

The ChimERA model can be seen as a proof of principle study of how a coupled fate and effect model

can be used as a realistic risk assessment tool incorporating more ecological and fate processes. The

uniqueness of the ChimERA model is that it used the input of various abiotic and biotic variables and

translated these into food web dynamics that varied greatly between the selected scenarios. This makes

models such as the ChimERA model ideally suited to address the question of multiple stressors (SCHER

et al. 2013; Gunderson et al. 2016): the ChimERA model provides an integrated prediction of all the

stressors present in the environment. Although the ChimERA model should be further improved with

e.g. additional environmental stressors (see 7.6), the integrated model (1) was able to provide realistic

effects of chemicals on food web dynamics based on the modelling of individual species, (2) included

spatiotemporal differences in exposure and effects on food web dynamics, (3) showed how the effects

of a chemical depend on the selected scenario, (4) included a feedback mechanisms between the abiotic

and biotic part of the system, (5) implemented an effect model capable of taking exposure history into

account and (6) considered the simultaneous effect of multiple chemicals.

(1) Realistic effects of chemicals on food web dynamics

Starting from the effects of a chemical assessed under laboratory conditions, the ChimERA model was

able to suggest how the chemical affects a hypothetical food web, taking into account both direct and

indirect effects. Indirect effects, occurring through competition and predation, need to be accounted for

in realistic ecological risk assessments (Fleeger et al. 2003). Indirect effects are very difficult to predict

based on single species toxicity tests alone and are difficult to account for in mesocosm studies (Van

den Brink and Ter Braak 1998). Ecological models have been suggested as good alternatives to account

for these indirect effects (Galic et al. 2010) and the ChimERA model demonstrated here how this can

be achieved: Asellus and Brachionus became dominant after chemical effects on Gammarus and

Daphnia, respectively, and the predator Chaoborus was negatively affected by chemical effects on its

prey Daphnia.

(2) Spatiotemporal differences in exposure and effects on food web dynamics

Although typically ignored in ecological risk assessments, the spatial structure of the system and the

location of emission can influence the effects of a chemical, e.g. as observed in mesocosms (Brock et

al. 2009) and predicted by models (Galic et al. 2012). Spatial structure will determine where and when

exposure is occurring (Wickwire et al. 2011), what fraction of the system is affected by the chemical

(Brock et al. 2009) and how fast the system can recover e.g. through immigration of biota from

unaffected locations (Dohmen et al. 2015). Because the spatial structure is explicitly taken into account

in the ChimERA model, the effects of the tested chemicals clearly differed between different locations

Chapter 6

118

in the system. The spatial dimension was not only important for differences in exposure. Differences in

food web structure in different parts of the system were also observed.. For example, as a result of

differences in movement potential of the different organisms, rotifers were more abundant and daphnids

less abundant in the streams than in the ponds.

(3) The outcome of chemical exposure depends on the selected scenario

Although the amount of chemical emitted to the system was identical in all scenarios, exposure

concentration and thus effects on the food web dynamics were highest in scenario III. The ChimERA

model was able to translate differences in environmental variables to differences in (1) food web

dynamics and (2) effects of chemical exposure. The three environmental scenarios studied here were

only a limited selection of the many possible combinations of temperature, trophic state and

hydrodynamics. Because of its flexibility, the ChimERA model structure can be tailored to specific real-

world situations by adjusting these environmental inputs. Similarly, the structure of the food web and

the species involved are exchangeable, thus allowing for adjustments to site-specific communities. The

scenarios studied here were limited to one spatial pattern i.e. two connected ponds. However, the

ChimERA model can be used to simulate scenarios for any kind of spatial structure and can thus be

applied to answer specific questions. For example, it could be used to assess how the presence of a non-

exposed part of the system affects the recovery of the food web.

The large number of possible scenarios can be confusing for risk assessors who need to decide e.g. if

the selected scenarios are applicable and/or if the food web and species are realistic. During the

development of Chimera, a similar problem was encountered for exposure assessment where there was

also a multitude of environmental fate models and scenarios available. To solve this, environmental

modelling was standardized and the FOCUS scenarios were developed (FOCUS 2001). A similar

standardization of ecological and effect models will help facilitate the acceptance of ecological models

as tools for effect assessment.

(4) Feedback mechanisms between the abiotic and biotic component of the system

The typical separate assessment of the fate and effects of chemicals ignores the fact that the

environmental fate is also influenced by the biological/ecological component of the system, and vice

versa. For example, the biomass of phytoplankton is an important factor for determining the bioavailable

concentration of a hydrophobic chemical (Morselli et al. 2015). However, phytoplankton is also a

component of the food web as it is consumed by grazers. Strong grazing pressure will limit the amount

of phytoplankton biomass in the system thus reducing the accumulation of the chemical in the

phytoplankton. The impact of this feedback mechanism was limited for the scenarios chosen here

because the effect of phytoplankton biomass on the chemical concentrations was limited. Other

biological compartments like macrophytes (Morselli et al. 2015) and zooplankton (Turner 2002) have

119

been shown to influence the environmental fate of a chemical significantly and can be included in future

scenarios.

(5) Effects of chemicals take earlier exposure into account

A TKTD model was used to predict the effects of chemicals on the survival of individuals. The nature

of this type of model allows effects to occur even when the environmental concentration of the chemical

have already decreased. This was visible e.g. for Daphnia exposed to chlorpyrifos in scenario II (Figure

6.8): at day 200, chlorpyrifos concentrations in the system had dropped to almost background

concentrations but adverse effects persisted and a population increase was only observed after day 210.

Although not tested here, TKTD models are especially useful when the emission is highly dynamic

(Jager et al. 2011). Even for chemicals that are emitted very irregularly, the TKTD approach is able to

predict the effects when only information of a typical toxicity test is available i.e. short term exposure

to a constant concentration.

(6) Simultaneous effects of multiple chemicals

The current implementation of the ChimERA model is able to predict the effects of a mixture of

chemicals, although the validity of the used mixture model is uncertain for the chemicals used in this

study. Mixture toxicity is complex and the selection of the appropriate mixture toxicity model depends

on the chemicals involved, although concentration addition was the preferred model in over 90% of

pesticide mixtures (Deneer 2000). The mechanisms of mixture toxicity between carbendazim,

chlorpyrifos and pyrene are unknown and an appropriate mixture toxicity model could therefore not be

chosen. The mixture toxicity model used here, although possibly invalid, did allow us to explore how

mixture toxicity effects may affect food web dynamics in comparison with single substance exposures.

Interestingly, different chemicals impacted different species: Daphnia was mainly affected by pyrene

while the effects of carbendazim and chlorpyrifos mainly affected Gammarus. Assessing the effects of

multiple chemicals on the food web is an important feature of the ChimERA model. Traditionally,

mixture toxicity studies focus on effects on the same organism. With the ChimERA model, mixture

toxicity effects can be evaluated for the whole food web. For example, how will the food web be affected

by the simultaneous exposure to a chemical affecting the grazers and one affecting the predators? The

ChimERA model can provide an answer to such a question by e.g. predicting that the negative effects

on the grazers are compensated by the reduced predation pressure, resulting in negligible effects of the

mixture on the grazer density.

Ecological risk assessment needs new methods to deal with the large amount of chemicals to which

ecosystems are possibly exposed and the increasing number of stressors that can interact with the effects

of a chemical (SCHER et al. 2013; Gunderson et al. 2016). Modelling has been named as one of the

most promising tools to face these challenges (SCHER et al. 2013). The simulations performed here

Chapter 6

120

with the ChimERA model are an innovative approach to risk assessment: environmental fate predictions

of the chemical are immediately coupled to its ecological effects and the effects could be differentiated

in space and time. However, models like the ChimERA model are so complex and offer so many options

that it can be difficult for risk assessors to correctly use them. How can such models be used as regulatory

instruments? Lessons can be learned here from how environmental fate models were accepted in Europe

i.e. the FOCUS scenarios and models (FOCUS 2001). Before FOCUS was defined, a multitude of

environmental fate models had been developed by scientists but risk assessors had no foundation to

evaluate these models and were reluctant to use them. Intensive collaboration between the scientific

community, industry and risk assessors led to the description of the FOCUS scenarios and a selection

of accepted FOCUS models for these scenarios. The FOCUS scenarios were selected to represent a

limited number of typical European situations such as a ditch system in The Netherlands or a typical

Scandinavian pond. The FOCUS models are environmental fate models that have been extensively

validated and are now accepted as ‘approved’ models to predict the environmental fate of a chemical in

the FOCUS scenarios. When a new chemical is developed, these FOCUS models can be used to predict

the environmental fate of the chemical in the FOCUS scenarios and thus be used in the exposure

assessment of the chemical. Because of the standardization, risk assessors can evaluate the application

of the FOCUS models, judge whether the environmental fate has been correctly assessed and ultimately

if the chemical poses a risk to the environment. Also, when new fate models are developed, these can

be tested with the FOCUS scenarios and compared with the validated FOCUS models. This allows new

models to be more easily accepted – and thus used – by regulators.

A similar standardization is required for ecological and toxicity models before these models will be

accepted by regulators and risk assessors. However, such standardization is hampered by the apparent

lack of universal ecological rules and the lack of clear protection goals (Van den Brink et al. 2006).

Question like what are acceptable effects in an ecosystem, should we focus on protecting biodiversity

or ecosystem functioning (or both?) and do we allow effects if recovery occurs (but within what time

frame?) need to be answered before standardized scenarios and models can be approved. This will

require a large effort from both the scientific community, industry and regulators but is, I believe,

absolutely necessary if we want ecological models and integrated models like the ChimERA model to

be accepted as risk assessment tools.

121

7 GENERAL CONCLUSIONS AND FUTURE

RESEARCH SUGGESTIONS

Chapter 7

122

7.1. Introduction

To perform adequate ecological risk assessment for the large amount of chemicals produced each year,

appropriate methods and procedures reflecting the environmental diversity and reality are needed.

Overall, current methods are, however, inadequate and lack ecological realism. New methods are

therefore needed and the use of ecological models has been suggested as a powerful tool to improve

ERA. In this PhD thesis the potential role of ecological interactions, one of the key characteristics

lacking from current ERA methods, was first studied experimentally. Individual based models (IBMs)

were then developed to simulate and evaluate the patterns observed in these experiments. Next, the

developed IBMs were integrated with an environmental fate model into the ChimERA model, capable

of simulating spatially-explicit effects of chemicals on a food web in realistic conditions. Finally, the

ChimERA model was tested as a scenario analysis tool.

The results of this PhD thesis are summarized in this chapter and suggestions for future research are

given. Each paragraph starts with the research question addressed in each chapter, then presents a

summary of the conclusions and finally suggests possible future research directions.

7.2. Species interactions and chemical stress: combined effects of intra- and

interspecific interactions and pyrene on Daphnia magna population

dynamics

The main research question addressed in Chapter 2 was how interactions within and between species –

intra- and interspecific competition with Brachionus calyciflorus and predation by Chaoborus sp. larvae

– influenced the effects of pyrene on populations of Daphnia magna.

Exposure to pyrene led to decreased D. magna densities and pyrene predominately affected smaller

individuals. Predation pressure by Chaoborus sp. larvae and intraspecific competition limited the D.

magna population densities but competition with B. calyciflorus had no significant effects on D. magna.

Predation pressure and intraspecific competition altered the size structure of the D. magna population,

reducing the fraction of small, more sensitive individuals in the population. As a result, predation and

intraspecific competition both interacted antagonistically with pyrene exposure i.e. reduced the effects

of pyrene. It can be concluded that, overall, interactions within and between species altered the effects

of pyrene, highlighting the need to account for such interactions in ecological risk assessments.

Future research suggestions: D. magna is a much stronger competitor and quickly outcompetes B.

calyciflorus in the used system. Probably, the influence of interspecific competition on the effects of a

chemical depends on the strength of the competition. How the strength of competition alters the effect

General conclusions and future research suggestions

123

of the chemical can be studied by e.g. using a smaller daphnid species. Similarly, the strength of the

predation pressure is expected to alter the effects of chemical exposure. This interaction can be studied

by using different predator species or densities. The effects of pyrene only became apparent after at least

one week of exposure. It would be interesting to see if more acute chemical effects (i.e. occurring sooner)

interact with species interactions in a different way. In the experiment, pyrene mainly affected the prey

(D. magna) and not the predator (Chaoborus sp. larvae). Applying a chemical that affects the predator

and not the prey can result in very different results.

The effects of a chemical are not only dependent on the ecological interactions studied in this chapter

i.e. competition and predation, but also on abiotic conditions. Multiple abiotic stressors can be present

in the environment and influence the toxicity of a chemical (SCHER et al. 2013; Gunderson et al. 2016).

Mixture toxicity of chemicals, for example, has received a lot of attention e.g. for metals (Jonker et al.

2005) and pesticides (Deneer 2000). Other, non-chemical stressors such as temperature (Scherer et al.

2013) can also greatly alter the effects of a chemical. Moreover, when and how often organisms are

exposed to these stressors is an important determining factor of their effects (Gunderson et al. 2016).

Chemical effects occurring in realistic conditions are thus much more complex than the controlled

environment studied in the laboratory. Ideally, future research should focus on studying how this

complex reality – including ecological interactions, multiple stressors and differences in timing –

influences the toxic effects of a chemical and how these can be accounted for in ecological risk

assessment. Alternative approaches to experimental work such as the models presented in this PhD

thesis are useful to address these challenges.

7.3. Development of the DEBkiss IBM

In Chapter 2, I have shown that interactions between species are essential to understand how chemicals

affect populations. The objective of Chapter 3 was to identify and develop a modelling approach that is

able to account for species interactions when assessing effects of chemicals.

Individual-based models (IBMs) are ideal tools to simulate effects of chemicals on populations. These

models, however, need to be based on a sound theoretical basis. Dynamic energy budget theory based

on the keep it simple, stupid principle (DEBkiss) offers a good compromise between complexity and

the amount of data required to parameterize such a model. DEBkiss IBMs were developed to account

for species interactions when species are exposed to chemicals. Two possible methods to calculate the

effects of a chemical were included: dose-response (DR) curves and toxicokinetic-toxicodynamic

models. The applicability of the DEBkiss IBMs and two toxicity model was evaluated in Chapter 4.

Chapter 7

124

7.4. Application of the DEBkiss IBM framework to assess the effects of

competition and chemical stress on the population dynamics of

Brachionus calyciflorus and Daphnia magna

In Chapter 4, I explored how the DEBkiss IBMs developed in Chapter 3 can be used to predict how

species interactions influence population responses to chemical exposure. Therefore, I applied the

DEBkiss IBMs developed in Chapter 3 to the experiments performed in Chapter 2.

DEBkiss IBMs were parameterized for D. magna and B. calyciflorus and compared to the experimental

results of Chapter 2. Population dynamics of isolated D. magna populations were reasonably well

predicted but the size structure of the population observed in Chapter 2 was not accurately predicted.

This was attributed to the absence of an energy reserve compartment in the models, leading to faster

starvation of the individuals when food is limited. The population dynamics of isolated B. calyciflorus

were accurately predicted using the DEBkiss IBM. Pyrene effects were predicted to occur too early with

both toxicity models but the TKTD model approached the observed effects best. The result of

competition was accurately predicted when both DEBkiss IBMs were coupled to a shared food source:

D. magna quickly outcompeted B. calyciflorus. Using the models to simulate competition when exposed

to pyrene resulted in an increase of B. calyciflorus when D. magna densities decreased. This was not

observed in the experiments and this can be explained by the fact that the predicted effects of pyrene

occurred earlier in the simulations than in the experiments.

Future research suggestions: the absence of a reserve compartment is proposed as the most likely

reason for the inability of the developed DEBkiss IBMs to predict the size structure of the D. magna

population. Application of an IBM based on the full DEB theory, where a reserve compartment is

included, would allow to evaluate this hypothesis. An IBM based on the full DEB theory is available

for D. magna (Martin et al. 2013a) and could be applied to these experiments to test whether this would

alleviate the problems observed with the DEBkiss IBM implementation. This full DEB IBM was

successfully applied to similar experiments with D. magna, although an additional starvation rule was

also required (Martin et al. 2013a). However, different D. magna clones can still differ significantly in

their parameters (Baird et al. 1991) and the experimental conditions were not identical e.g. different

algal food source and feeding regime. It is therefore unsure whether this DEB IBM could be applied

without adjustments to the experiments of Chapter 2.

Ideally, the DEBkiss IBM implementation needs to be validated further. The models need to be

compared with experiments where the same species were used in other conditions e.g. different food

concentrations, different starting densities or exposure to different chemicals. This would greatly

increase the acceptance of these models. Also, application of DEBkiss IBMs to other combinations of


125

species where e.g. the strength of competition is different, would show the general applicability of

DEBkiss IBMs to other species. The applicability for different types of species interactions (e.g.

predation) could also be explored further.

7.5. Development of the integrated ChimERA model

The objective of Chapter 5 was to develop an integrated ecological risk assessment model. As a first

step, a food web model was described based on the DEBkiss IBMs from Chapter 3. To provide realistic

simulations of the effects of a chemical on food web dynamics, the food web model was integrated with

an environmental fate model.

The ChimERA model was developed as a tool to perform more realistic ecological risk assessments.

The ChimERA model integrates an environmental fate model (ChimERAfate) and a food web model

based on DEBkiss IBMs (ChimERAfoodweb). ChimERAfate is a dynamic and spatially explicit fate model

that predicts environmental concentrations based on environmental variables (hydrodynamics,

temperature and trophic state). ChimERAfoodweb was developed based on the earlier implementation of

competition between DEBkiss IBMs and a newly implemented predation interaction. The food web

model included Brachionus and Daphnia as grazers, Chaoborus as a predator and Asellus and

Gammarus as detritus feeders. The ChimERA model was used in Chapter 6 to evaluate the effects of

chemicals on food web dynamics for different hypothetical scenarios.

7.6. The ChimERA model as a scenario analysis tool

Models have been suggested as good alternatives for current risk assessment methods. In Chapter 6, I

explored how the developed ChimERA model could be used to predict the risk of chemicals for a

range of different environmental conditions.

The ChimERA model was applied to a two-pond system in 15 hypothetical scenarios, differing in water

residence time, temperature, trophic state and the applied chemical. In general, the ChimERA model

predicted food web dynamics and effects of chemicals that varied greatly between the 15 scenarios

considered. Densities of the species were highest in scenarios with a high trophic state, high temperature

and high water residence time. Daphnids were the dominant grazers in the ponds while rotifers

dominated in the streams. Gammarus was the dominant detritus feeder. Concentrations of applied

chemicals (carbendazim, chlorpyrifos and pyrene) were highest in scenarios with a high water residence

time. The physico-chemical properties of the chemical determined the spatio-temporal pattern of the

exposure, i.e. where the concentrations were highest and how fast the chemical disappeared. Direct

effects of the chemicals on Chaoborus, Daphnia and Gammarus densities were predicted, as expected

from their sensitivities to these chemicals. These effects were, however, heterogeneously distributed in

Chapter 7

126

space and time, following the differences in exposure. The most notable indirect effects were a shift in

dominance from Daphnia and Gammarus to Brachionus and Asellus, respectively. Also, the predator

Chaoborus was affected indirectly through effects on its prey species Daphnia. The effects of the

mixture of three chemicals differed between species, where the effects of pyrene dominated for Daphnia

and the effects of carbendazim and chlorpyrifos for Gammarus.

Future research suggestions: although the current work gives a good overview of the potential use of

an integrated fate and effect model for risk assessment, further steps are needed before the model can be

applied as a risk assessment tool. The suggested steps can be divided in three categories: more

complexity, more testing and guidance on the model output.

(1) More complexity

Multiple stressors and the timing of exposure to these stressors have been raised as a critical points

missing in current ecological risk assessment (SCHER et al. 2013; Gunderson et al. 2016). The

ChimERA model presented in this thesis offers a good first step on how these can be accounted for.

However, although the ChimERA model is already able to account for realistic exposure scenarios, the

complexity of the submodels is limited and probably needs to be increased further to better simulate

reality. Some key areas where complexity can be improved are discussed here.

The DEBkiss IBMs currently used in the food web model are most likely insufficient to accurately

simulate the full life cycles of Asellus, Chaoborus, Gammarus and possibly also Daphnia. DEBkiss

IBMs were used here because of their transparency and simplicity. This allowed for an easy

interpretation of the food web dynamics and resulted in reasonable the computation times. Although

their parametrization was based on a verified database and literature sources, the DEBkiss

implementation has not been tested extensively against actual observational data. However, given the

problems with the DEBkiss implementation for Daphnia (Chapter 4), it is likely that similar problems

with e.g. the absence of a reserve compartment would arise for other species. The ChimERA framework

was developed as a very flexible tool and, as long as the same information is exchanged with the server,

other IBMs can be implemented in the food web. However, a tested implementation of the full DEB

theory into an IBM is only available for Daphnia (Martin et al. 2012). Adaptation and validation of this

implementation for the other species was not in the scope of this research. Validated IBMs not based on

DEB theory are available for Asellus (Van den Brink et al. 2007), Chaoborus (Strauss et al. 2016) and

Gammarus (Galic et al. 2014). However, these implementation do not always explicitly consider food

and the use of different underlying theories could hinder interpretation.

The complexity of several ecological processes included in the ChimERA model can be improved, of

which a few will be discussed here in detail. One example of such an ecological process deals with the

movement of individuals within the spatial structure. In the current simulations, random movement was

assumed and this was shown to cause mobile individuals to accumulate in the ponds. However,


127

movement of individuals is not random and is dependent on the prevalent conditions. For example, the

spatial distribution of Gammarus is highly dependent on the stream conditions (Adams et al. 1987),

Chaoborus larvae move to other regions when prey become limited (Liljendahl-Nurminen et al. 2002)

and Daphnia longispina are known to actively avoid contaminant exposure (Lopes et al. 2004).

Movement within the system is thus dependent on many species-specific, often unknown, factors ,

making the modelling of realistic movement very complex and not within the scope of this PhD thesis.

However, it could be worthwhile to explore this for species for which there is a known relationship

between movement and an environmental factor.

The selection and quality of food is another example of complexity that could be added. Food in the

current implementation of the ChimERA model is limited to a very general description of phytoplankton

(or detritus). However, in reality, phytoplankton consists of many species belonging to different algal

classes. Grazers prefer to eat certain phytoplankton species and phytoplankton species differ in their

nutritional value for the grazer. It is likely that differences in food selectivity and food quality

(nutritional value) will impact the energy budgets of the individual species. Modelling multi-species

grazing e.g. similar to how multi-species predation was added to the ChimERA model would allow to

explore the implications of increasing the complexity of the phytoplankton compartment. Similarly, the

detritus consists of many fractions, also from non-algal origin, and accounting for this will increase the

realism of the models.

Similarly, increasing the complexity of the fate model can be considered by including additional

parameters and processes of importance. The fate model at the moment does not consider high frequency

fluctuations in water chemistry parameters that can be important for the environmental fate of the

chemical. For example, pH influences several environmental fate processes such as adsorption to soil

and degradation (Vala Ragnarsdottir 2000) but high frequency (e.g. day-night) fluctuations are not

included in ChimERAfate. Inclusion of such processes will further increase the accuracy of the fate

model. Another important improvement to the chemical fate model would be the addition of chemical

effects on the phytoplankton dynamics. For the three chemicals studied, phytoplankton effects were

considered negligible. However, for other chemicals such as herbicides, effects on phytoplankton are

likely. Because phytoplankton dynamics in the fate model are modelled using an ordinary differential

equation (Equation 5.1), the effects of chemicals could be included easily by multiplying the

phytoplankton growth rate with a concentration-response term (De Laender et al. 2015).

Further development of the ChimERA model, or similar models, should strive to increase the spatial

scale and the simulated period. Current application was limited to two connected ponds simulated for

one year. Increasing the spatial scale and the simulated period will provide more in-depth, realistic

ecological risk assessments. Questions such as what are the multi-annual effects of a chemical? Or how

will a different spatial scale and structure affect the effects of a chemical e.g. when multiple isolated

Chapter 7

128

ponds are located within a larger area? could be addressed. Also, new methods in ecology could be used

to further increase the relevance of the risk assessment. For example, geographic information system

(GIS) approaches have become popular in (landscape) ecology e.g. for fire risk assessment (Chuvieco

et al. 2010). Combining a GIS approach with the ChimERA model could lead to ecological risk

assessments for real-world situations.

Finally, it would be interesting to explore the effects of chemicals for other food web types. The current

ChimERAfoodweb model only considers five species and three trophic levels (phytoplankton/detritus,

grazer/detritivore and a predator). Both direct and indirect effects of a chemical could be totally different

when e.g. there are more trophic levels and more trophic links in the food web. Comparing effects

between these food webs will allow more insight into how chemicals can affect food web dynamics.

(2) More testing

Parts of the ChimERA model were tested: e.g. DEBkiss IBMs for two competing grazers (see Chapter

4) and the ChimERAfate model (Morselli et al. 2015). However, the applicability of ChimERA for more

complex cases needs to be tested further by comparing the simulations with observational data. The

DEBkiss IBMs have only been applied for species with a short life cycle. The applicability for species

that live multiple months and even years needs to be further tested. Also, the DEBkiss IBMs were only

compared with data for two interacting species (Chapter 4). In real food webs, interactions occur

between more than two species. To test how well more complex species interactions are modelled,

DEBkiss IBMs could be compared to the results of experiments in the lab with simplified food webs,

e.g. including multiple grazers and predators.

(3) Guidance on the model output

The interpretation of the output of complex models such as the ChimERA model is a challenge. For

non-experts, the model output can be intimidating and confusing. To increase the acceptance of the

ChimERA model, clear guidance needs to be provided on how the model works and how the output

should be interpreted. For the ChimERA model to be used as a risk assessment tool, further definition

is required of which endpoints should be used and what effects are considered acceptable. This is of

course strongly dependent on what regulators require and input from them is therefore pivotal at this

point. Additionally, good modelling practice and standardized model documentation protocols such as

TRACE (Schmolke et al. 2010) are essential in this regard to increase the transparency and general

acceptance of models in general and the ChimERA model specifically.


129

7.7. Overall contribution of this PhD thesis to the future of ecological risk

assessment

This work has shown the importance of approaching chemical risk assessments from an ecological

perspective. Species are not isolated entities but interact and influence the outcome of chemical

exposure. The ChimERA model developed and applied here is an important proof of principle on how

these interactions can be accounted for. In addition, the model also accounted for spatiotemporal

differences in exposure and effects of chemicals. The ChimERA model, when further tested, validated

and extended with more mechanisms, can serve as a blueprint for future efforts to model how chemicals

affect ecosystems. The integration of a fate and effect model demonstrated the strength of

simultaneously accounting for both the exposure and effects of chemicals when assessing potential

ecological risks.

131

A

SUPPORTIVE INFORMATION CHAPTER 2

Appendix A

132

Table A1: Percentage of the total variance explained by pyrene exposure and intraspecific competition. Percentages shown are calculated for the optimal GLM for log10-transformed total D. magna abundance after backwards model selection on the intraspecific dataset. Non-significant predictor variables are indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 Pyrene / 13.7 10.4 2.1 16.6 12.3 75.4 32.1 Intraspecific 71.0 46.7 21.9 20.4 40.3 38.5 0.9 21.6 Pyrene X Intraspecific / / / 23.2 13.9 / 8.4 /

Table A2: Percentage of the total variance explained by of pyrene exposure and interspecific competition. Percentages shown are calculated for the optimal GLM for log10-transformed total D. magna abundance after backwards model selection on the interspecific dataset. Non-significant predictor variables are indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 Pyrene / 19.5 / / 5.1 23.1 81.3 24.0 Interspecific / 12.0 / 51.8 55.8 32.1 4.0 14.0 Pyrene X Interspecific / / / / 9.6 / / 18.2

Table A3: Percentage of the total variance explained by of pyrene exposure and predation. Percentages shown are calculated for the optimal GLM for log10-transformed total D. magna abundance after backwards model selection on the predation dataset. Non-significant predictor variables are indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 Pyrene / / / / / / 45.1 13.3 Predation / 11.4 51.8 42.4 46.8 54.6 19.5 55.0 Pyrene X Predation / / / / / / 16.8 /

Table A4: GLM estimates of pyrene exposure and intraspecific competition for log10-transformed adult D. magna abundance after backwards model selection. For each time point, the significant estimates (p< 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variablesare not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) 0.57 1.02 1.11 1.22 1.40 1.48 1.69 1.73 Medium pyrene / / / / / 0.10 / / High pyrene 0.12 / / / -0.12 / -0.37 -0.18Low intraspecific 0.30 0.24 0.19 0.16 0.08 / / / High intraspecific 0.46 0.41 0.37 0.32 0.25 0.08 / /

Supportive Information Chapter 2

133

Table A5: GLM estimates of pyrene exposure and intraspecific competition for log10-transformed juvenile D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) 0.74 0.56 1.15 1.57 1.89 1.79 1.56 1.08 Low pyrene / / / / -0.17 / / 0.44 High pyrene / / / / / / -0.70 / Low intraspecific 0.22 / / / -0.24 -0.14 / / High intraspecific 0.49 0.57 0.27 -0.21 -0.33 -0.28 / / Low pyrene X Low intraspecific / / / / 0.32 / / / Medium pyrene X High intraspecific / / / 0.28 / / / -0.63

Table A6: GLM estimates of pyrene exposure and intraspecific competition for log10-transformed neonate D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) / / 1.25 1.47 1.47 0.69 / -0.52 Medium pyrene / 0.79 / / / 0.32 / / High pyrene / 0.87 / / / 0.33 / 1.03 Low intraspecific 0.51 / / -0.39 -0.34 / / / High intraspecific 0.75 / -0.71 -0.75 -0.61 -0.55 / / High pyrene X Low intraspecific / -1.37 / / / / / / High pyrene X High intraspecific / -1.02 / / / / / /

Table A7: GLM estimates of pyrene exposure and interspecific competition for log10-transformed adult D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) 0.61 1.03 1.14 1.20 1.42 1.49 1.72 1.77 Medium pyrene / / / / / 0.10 / / High pyrene / / / / -0.09 -0.10 -0.57 -0.22 Low interspecific / / / 0.13 / / / / High pyrene X High interspecific / / / / / / / -0.29

Appendix A

134

Table A8: GLM estimates of pyrene exposure and interspecific competition for log10-transformed juvenile D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) 0.79 0.56 1.21 1.56 1.89 1.83 1.59 1.06 Low pyrene / / / / -0.17 / / / High pyrene / / / / / -0.14 -0.81 0.42 Low interspecific / 0.35 / / / -0.15 -0.32 -0.42 High interspecific -0.35 0.32 / -0.14 -0.27 -0.16 -0.23 / Low pyrene X Low interspecific / / / / 0.22 / / / Low pyrene X High interspecific / / / / 0.23 / / / High pyrene X High interspecific 0.59 / / / / / / /

Table A9: GLM estimates of pyrene exposure and interspecific competition for log10-transformed neonate D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) -0.48 / 1.15 1.44 1.52 0.57 / -0.81 Medium pyrene 0.56 / / / / 0.56 / / High pyrene / / / / / / / 1.79 Low interspecific -0.59 / / -0.38 -0.49 / / / High interspecific / / / -0.41 -0.57 -1.22 / / High pyrene X Low interspecific / -1.58 / / / / / / Low pyrene X High interspecific / / / / / 1.21 / / High pyrene X High interspecific / / / -0.44 / 1.00 / /

Table A10: GLM estimates of pyrene exposure and predation for log10-transformed adult D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) 0.67 1.02 1.11 1.22 1.36 1.49 1.67 1.76 High pyrene / / / / / / -0.36 -0.16 Predation / -0.20 -0.22 -0.22 -0.18 / / -0.16


135

Table A11: GLM estimates of pyrene exposure and predation for log10-transformed juvenile D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) 0.67 0.55 1.01 1.56 1.89 1.79 1.56 1.27 Medium pyrene / / / / -0.28 / / / High pyrene / / / / / / -0.70 / Predation / / / -0.34 -0.36 -0.37 -0.50 -0.33

Table A12: GLM estimates of pyrene exposure and predation for log10-transformed neonate D. magna abundance after backwards model selection. For each time point, the significant estimates (p < 0.05) of explanatory variables and their interactions are shown. Non-significant predictor variables are not shown (if never significant) or indicated with “/”.

Time (days) -4 0 2 4 7 10 15 22 (Intercept) / / 1.25 1.51 1.47 0.57 -0.42 -0.57 High pyrene / 0.79 / / / / / 0.96 Predation / / -0.67 -1.03 -0.80 / / / Low pyrene X Predation / / / / / -1.07 / /

Appendix A

136

Figure A1. Optimal concentration response curve for pyrene fitted to the results of a 48 hours test with neonate D. magna.

Figure A2. Adult D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the adult D. magna abundances with no additional species interactions. Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.


137

Figure A3. Adult D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments without (points) and with predation (crosses). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Figure A4. Adult D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments with no additional species interactions (points), low intraspecific competition (crosses) and high intraspecific competition (black squares). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Appendix A

138

Figure A5. Adult D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments with no additional species interactions (points), low interspecific competition (crosses) and high interspecific competition (black squares). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Figure A6. Juvenile D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the juvenile D. magna abundances with no additional species interactions. Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.


139

Figure A7. Juvenile D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments without (points) and with predation (crosses). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Figure A8. Juvenile D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments with no additional species interactions (points), low intraspecific competition (crosses) and high intraspecific competition (black squares). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Appendix A

140

Figure A9. Juvenile D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments with no additional species interactions (points), low interspecific competition (crosses) and high interspecific competition (black squares). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Figure A10. Neonate D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the neonate D. magna abundances with no additional species interactions. Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.


141

Figure A11. Neonate D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments without (points) and with predation (crosses). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Figure A12. Neonate D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments with no additional species interactions (points), low intraspecific competition (crosses) and high intraspecific competition (black squares). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Appendix A

142

Figure A13. Neonate D. magna abundance over time for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Data shown are the treatments with no additional species interactions (points), low interspecific competition (crosses) and high interspecific competition (black squares). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the first and the second pyrene application.

Figure A14. B. calyciflorus population sizes over time with D. magna present for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Populations were started with either 333 rotifers · vessel-1 (points) or 999 rotifers · vessel-1 (crosses). Average values with standard deviations (error bars) are depicted. Dashed lines indicate first and the second pyrene application.


143

Figure A15. B. calyciflorus population sizes over time with D. magna present for four pyrene exposure profiles: control, low, medium and high pyrene exposure. Populations were started with either 333 rotifers · vessel-1 (points) or 999 rotifers · vessel-1 (crosses). Average values with standard deviations (error bars) are depicted. Dashed lines indicate first and the second pyrene application.

Figure A16. B. calyciflorus population sizes over time without competition with D. magna and for two pyrene exposure profiles: no exposure (“Control”) and high exposure (110 μg/L, “C4”). Average values with standard deviations (error bars) are depicted. Dashed lines indicate the time of pyrene application.

145

B SUPPORTIVE INFORMATION CHAPTER 4

Appendix B

146

Figure B1: Likelihood (LSS) of the DEBkiss models with values for the maximum area-specific assimilation rate (max_spec_assimr) between 0.006 and 0.121 mg mm-2 d-1.


147

Figure B2: Total population density (A) and densities of large (B), medium (C) and small (D)

individuals for D. magna using a DEBkiss IBM without additional food starvation mortality. Black

bullets and squares show the observed dynamics for populations without additional stress in two

experiments. Green areas show the best predictions.

Appendix B

148

Figure B3: Total population density (A) and densities of large (B), medium (C) and small (D) individuals for D. magna using a DEBkiss IBM with additional food starvation mortality. Black bullets and squares show the observed dynamics for populations without additional stress in two experiments. Green areas show the best predictions.

149

C SUPPORTIVE INFORMATION CHAPTER 5

Appendix C

150

C.1. Literature sources for DEBkiss parameters

Asellus: (Maltby 1991, 1995; Arakelova 2001; Galic et al. 2012)

Brachionus: (Halbach 1970a; Dumont et al. 1975; Saunders III and Lewis Jr 1988; Hansen and Bjornsen

1997; Mohr and Adrian 2000; Jensen and Verschoor 2004)

Chaoborus: (Stenson 1978; Van Wijngaarden et al. 2006)

Daphnia: (MacArthur and Baillie 1929; Dumont et al. 1975; Tillmann and Lampert 1984; Ebert 1992;

Glazier 1992; Trubetskova and Lampert 1995; Jager and Zimmer 2012; Mulder and Hendriks

2014)

Gammarus: (Nilsson 1977; Welton and Clarke 1980; Gee 1988; Ward 1988; McCahon and Pascoe

1990; Maltby 1995)

C.2. Derivation of multi-species predation equation

Disc equation for multiple prey:

(Rose et al 1999)

Where Cj = the biomass of prey j eaten by predator I (g); Cmax = the maximum food uptake by predator

i (g d-1); Wi = the weight of the predator i (g); PDj = the density of prey j (prey L-1); Kj = the half-

saturation coefficient for prey j (prey L-1); n = the numer of prey species for predator i.

Data available for Daphnia and Brachionus are based on:

′

′ (Krylov 1992)

Where PR = predation rate (prey predator-1 d-1); a’j = attack rate on prey j (L d-1); Nj = density of prey j

(prey L-1); Thj = handling time of prey j (d); n = the number of prey species.

To implement multi-species predation in the IBMs, we need the biomass of prey eaten. However, we

only have the parameters for the Krylov 1992 equation. Therefore, we combine both equations using the

half-saturation coefficient K. The half-saturation coefficient K in a type II functional response function

corresponds with the food density at which the food uptake is exactly 50% of the maximum food uptake.

For the Krylov 1992 equation, the maximum food uptake corresponds to the limit of the function i.e.

. Therefore the Krylov 1992 equivalent of K is equal to . The corresponding food density N is

calculated by:


151

′

′ => ′

Combined with the Disc equation, the amount of prey biomass Cj eaten can be calculated:

with ′

′

C.3. Communication protocol

Communication between components in the ChimERA model

Two considerations were taken into account for the design of the communication between the different

model programs via the server. Firstly, without a minimum of synchronization imposed upon the system

by the server program, it will be very hard to get a robust system. Synchronization implies here that all

models/clients should receive a signal to perform their computations and all models/clients inform the

server program when they are ready. Only when all model/clients are ready, the system should proceed

with the next step. Secondly, the different types of data that need to be exchanged require a two-step

process. Some data are state data that can be exchanged as soon as all (new) states are calculated. Other

data however might refer to fluxes between models that are calculated only within one model but need

to be accounted for in other models before their new state can be calculated. These data can only be send

to model 2 when calculations in model 1 have been done. The two-step process thus involves first

calculating changes in all models, then exchanging deltas (loss or increase terms) and then calculating

the final next state.

Main structure

With the generic architecture presented here, the complete system can be split up in an arbitrary number

of modules. Models have an associated client, that handles communication with the server. For the

ChimERA model, one client handles the communication with the fate model, one client the

communication with the two detritus feeders and one client the communication with the grazers and

their predator (Figure C1).

Appendix C

152

Figure C1: Setup of the integrated ChimERA model. The server communicates with three clients. One client coordinates the chimERA fate model and the two other clients coordinate the DEBkiss IBMs: one client for the two detritus feeders (ASE and GAM for Asellus and Gammarus, respectively) and one client for the two grazers and the predator (BRA. DAP and CHA for Brachionus, Daphnia and Chaoborus, respectively).

Table C1: Data exchanged between the communication server and clients.

Name Type Definition Unit

PHY STATE Phytoplankton concentration mg ww L-1

CPW STATE Concentration chemical in water mol L-1

PGL DELTA Loss (flux) of phytoplankton due to grazing mg ww L-1 h-1

DET STATE Detritus (in water) concentration mg ww L-1

DGL DELTA Loss (flux) of detritus due to grazing mg ww L-1 h-1

Communication at connection time

When first connecting with the sever, clients need to specify which data it needs to receive,

distinguishing between state variables and fluxes. The two types are indicated by &STATE and

&DELTA, and should be followed by the character @ and zero or more (comma-separated) keywords

for the data, e.g., '&STATE@PHY,ZOOP1,ZOOP2&DELTA@GRA*'. Note that it should be possible

to include models that do not need input from other models, so signalling: '&STATE@&DELTA@*' or

just one of the two types: '&STATE@&DELTA@GRA*'

Communication during simulation


153

The basic loop for each time step is worked out below. The following synchronized steps occur, starting

at the moment all models’ state variables have just been updated. Exclamation marks are used to indicate

the start of a command given by the server, or the following reply coming from the client.

1. Server sends: !calcDelta* to all (waiting) clients and waits for replies from all. No data accompanies

this command. The client allows its model to run up to the moment the next/new state is calculated.

It has to stop then, because the calculation of the final new state may require information exchange

with other models.

2. All clients send: !delta* to the server, to signal that they are ready. When a client produced

information that is needed in other models, the standard data string (possibly composed) with this

information should be appended to !delta, e.g. !delta#1&GRA@1,2,3,4,5,6,7,8,9,10*. After sending

confirmation/data, clients wait.

3. When all confirmations have arrived at the server, the server starts the exchange of delta data by

building for all clients that expressed the need for these data, the pure (and possibly composed) delta

data strings, and subsequently sending them. The information from the server has the format

!delta<data-string>*, with data-string in the format we decided upon, e.g.

!delta#1&PGL@1,2,3,4,5,6,7,8,9,10&DGL@9,8,7,6,5,4,3,2,1,0* or

!delta#1&PGL@1,2,3,4,5,6,7,8,9,10*. In the data string, any arbitrary data can be included. These

data – not needed by other models – will not be exchanged between component models, but only

stored and visualized by the server. E.g.,

!delta#1&PGL@1,2,3,4,5,6,7,8,9,10&MYDATA@9,8,7,6,5,4,3,2,1,0*.

4. Server waits until all the clients that needed delta data have signalled that these data have been

received, e.g. by replying !received*

5. All clients now wait until the server sends them the message !calcState*. Upon receiving this

message the clients make their model do the necessary thing to set the new state of the model (e.g.

subtracting or adding fluxes calculated in other compartments/models).

6. When new state has been set in the models, all clients have to signal that they are ready, by sending

!state* to the server. In case clients provide state data that are needing in other models, the (possibly

composed) state data are appended: !state<data-string>* e.g. !state#1&PHY@1,2,3,4,5,6,7,8,9,10*.

After sending confirmation, clients wait. Also here, in the data string, any arbitrary data can be

included. These data – not needed by other models – will not be exchanged between component

models, but only stored and visualized by the server. E.g.,

!state#1&PHY@1,2,3,4,5,6,7,8,9,10&MYDATA@9,8,7,6,5,4,3,2,1,0*.

7. When all these confirmations have arrived at the server, the server starts the exchange of state data

by building for all clients that expressed the need for these data, the pure (and possibly composed)

state data strings, and subsequently sending them.

Appendix C

154

8. Server waits until all the clients that needed state data have signalled that these data have been

received, e.g. by replying !received*

9. Etc. continuing with step 1. This is the moment the server might be interrupted, leaving the system

in a resumable state.

Some specials steps need to be taken when initializing the whole system. Initialization can be done by

a variant of the steps 5 to 8 (related to setting state):

10. After the connections have been made (open socket connection created between server and client)

all clients wait until the server sends them the message !initialize*. Upon receiving this message the

clients make their model do the necessary initializations to set the state of the model at time 0 (this

refers to state that can be initialized without information on the initial state of other models).

11. When initializations are done in the models, all clients have to signal that they are ready, by sending

!initialized* to the server. In case clients provide state data that are needed in other models, the

(possibly composed) state data are appended, e.g., !initialized#1&PHY@1,2,3,4,5,6,7,8,9,10*.

After sending confirmation, clients wait. (we could also just copy/reuse the !state* protocol).

12. When all these confirmations have arrived at the server, the server starts the exchange of state data

by building for all clients that expressed the need for these data, the pure (and possibly composed)

state data strings, and subsequently sending them.

13. Server waits until all the clients that needed state data have signalled that these data have been

received, e.g. by replying !received*. After this, all models have their complete up to date state

related to t = 0.

155

D SUPPORTIVE INFORMATION CHAPTER 6

Appendix D

156

Table D1: Physico-chemical properties of the three selected chemicals. MW = molecular weight; VP = vapour pressure; WS = water solubility; Kow = octanol-water partition coefficient; HLwater = half-life in water.

Chemical MW (g mol-1) VP (Pa) WS (mg L-1) Log KOW HLwater (h)

Carbendazim 191.2 6.48 * 10-8 8 1.52 720

Chlorpyrifos 350.6 2.27 *10-3 0.73 4.96 24

Pyrene 202.3 6 * 10-4 0.132 5.18 1700

Table D2: LC50 values and TKTD parameters for the five species considered in the food web. LC50 values are reported in μg/L. kD = dominant rate constant (d-1); kK = killing rate constant (L μg-1 d-1); z = internal threshold for effects (μg L-1); ns = Not sensitive at the concentrations tested.

Parameter Asellus Brachionus Chaoborus Daphnia Gammarus

Carbendazim

EC50 350 ns ns 91 55

kD 4.6 * 10-4 1 * 10-6 1 * 10-6 2.5 * 10-3 0.05

kK 4.7 * 10-3 1 * 10-6 1 * 10-6 0.46 0.021

z 6.7 1 * 105 1 * 105 0.3 7.5

Chlorpyrifos

EC50 8.58 ns 0.3 0.82 0.23

kD 9 * 10-4 1 * 10-6 53 0.3 0.7

kK 27 1 * 10-6 0.23 0.35 2

z 0.003 1 * 105 4.4 * 10-4 6.1 * 10-8 0.006

Pyrene

EC50 303.5 ns ns 68 27.1

kD 0.021 1 * 10-6 1 * 10-6 0.969 0.23

kK 0.022 1 * 10-6 1 * 10-6 0.008 0.11

z 3.5 1 * 105 1 * 105 0 8.4

157

REFERENCES

References

158

Abbriano RM, Carranza MM, Hogle SL, Levin RA, Netburn AN, Seto KL, et al. Deepwater Horizon Oil Spill: A Review of the Planktonic Response. Oceanography. 2011;24(3):294–301.

Adams J, Gee J, Greenwood P, McKelvey S, Perry R. Factors affecting the microdistribution of Gammarus pulex (Amphipoda): an experimental study. Freshw Biol. 1987;17(2):307–16.

Arakelova KS. The evaluation of individual production and scope for growth in aquatic sow bugs (Asellus aquaticus). Aquat Ecol. 2001;35(1):31–42.

Augusiak J, Van den Brink PJ, Grimm V. Merging validation and evaluation of ecological models to “evaludation”: A review of terminology and a practical approach. Ecol Modell. 2014;280:117–28.

Baird DJ, Barber I, Bradley M, Soares AMVM, Calow P. A comparative study of genotype sensitivity to acute toxic stress using clones of Daphnia magna straus. Ecotoxicol Environ Saf. 1991;21(3):257–65.

Baird DJ, Maltby LL, Greig-Smith PW, Douben PET. ECOtoxicology: Ecological dimensions. Chapman & Hall Ecotoxicology Series; 1996.

Beketov M a, Liess M. The influence of predation on the chronic response of Artemia sp. populations to a toxicant. J Appl Ecol. 2006;43(6):1069–74.

Beketov MA, Liess M. Ecotoxicology and macroecology - Time for integration. Environ Pollut. 2012;162:247–54.

Bellas J, Saco-Alvarez L, Nieto O, Beiras R. Ecotoxicological evaluation of polycyclic aromatic hydrocarbons using marine invertebrate embryo-larval bioassays. Mar Pollut Bull. 2008;57(6-12):493–502.

Bontje D, Kooi BW, Liebig M, Kooijman SALM. Modelling long-term ecotoxicological effects on an algal population under dynamic nutrient stress. Water Res. 2009;43(13):3292–300.

Bradbury SP, Feijtel TCJ, Van Leeuwen CJ. Meeting the scientific needs of ecological risk assessment in a regulatory context. Environ Sci Technol. 2004;38(23):463a – 470a.

Bray JR, Curtis JT. An Ordination of the upland forest community of southern Wisconsin.pdf. Ecol Monogr. 1957;27(4):325–49.

Bridges CM. Long-term effects of pesticide exposure at various life stages of the southern leopard frog (Rana sphenocephala). Arch Environ Contam Toxicol. 2000;39(1):91–6.

Brock TCM, Arts GHP, Maltby L, Van den Brink PJ, Maltby ÀL, Brink PJ Van Den. Aquatic risks of pesticides, ecological protection goals, and common aims in european union legislation. Integr Environ Assess Manag. 2006;2(4):e20–46.

Brock TCM, Roessink I, Belgers JDM, Bransen F, Maund SJ. Impact of a Benzoyl Urea Insecticide on Aquatic Macroinvertebrates in Ditch Mesocosms with and without Non-Sprayed Sections. Environ Toxicol Chem. 2009;28(10):2191–205.

Cairns J. Putting the eco in ecotoxicology. Regul Toxicol Pharmacol. 1988;8(2):226–38.

Caquet T, Deydier-Stephan L, Lacroix G, Le Rouzic B, Lescher-Moutoue F. Effects of fomesafen, alone and in combination with an adjuvant, on plankton communities in freshwater outdoor pond mesocosms. Environ Toxicol Chem. 2005;24(5):1116–24.

Cardinale BJ, Duffy JE, Gonzalez A, Hooper DU, Perrings C, Venail P, et al. Biodiversity loss and its impact on humanity. Nature. 2012;486(7401):59–67.

References

159

Chapman PM, Fairbrother A, Brown D. A critical evaluation of safety (uncertainty) factors for ecological risk assessment. Environ Toxicol Chem. 1998;17(1):99–108.

Chapman PM. Integrating toxicology and ecology: putting the “eco” into ecotoxicology. Mar Pollut Bull. 2002;44(1):7–15.

Chuvieco E, Aguado I, Yebra M, Nieto H, Salas J, Martín MP, et al. Development of a framework for fire risk assessment using remote sensing and geographic information system technologies. Ecol Modell. 2010;221(1):46–58.

Clements WH, Rohr JR. Community responses to contaminants: Using basic ecological principles to predict ecotoxicological effects. Environ Toxicol Chem. 2009;28(9):1789–800.

Coors A, Meester L De. Synergistic, antagonistic and additive effects of multiple stressors: predation threat, parasitism and pesticide exposure in Daphnia magna. J Appl Ecol. 2008;45:1820–8.

Daam MA, Van den Brink PJ. Implications of differences between temperate and tropical freshwater ecosystems for the ecological risk assessment of pesticides. Ecotoxicology. 2010;19(1):24–37.

Day KE, Kaushik NK, Solomon KR. Impact of Fenvalerate on Enclosed Freshwater Planktonic Communities and on in situ Rates of Filtration of Zooplankton. Can J Fish Aquat Sci. 1987;44(10):1714–28.

De Hoop L, De Troch M, Hendriks a J, De Laender F. Modeling toxic stress by atrazine in a marine consumer-resource system. Environ Toxicol Chem. 2013;32(5):1088–95.

De Laender F, De Schamphelaere KAC, Vanrolleghem P a, Janssen CR. Is ecosystem structure the target of concern in ecological effect assessments? Water Res. 2008a;42(10-11):2395–402.

De Laender F, De Schamphelaere KAC, Vanrolleghem P a, Janssen CR. Validation of an ecosystem modelling approach as a tool for ecological effect assessments. Chemosphere. 2008b;71(3):529–45.

De Laender F, De Schamphelaere KAC, Vanrolleghem PA, Janssen CR. Do we have to incorporate ecological interactions in the sensitivity assessment of ecosystems? An examination of a theoretical assumption underlying species sensitivity distribution models. Environ Int. 2008c;34(3):390–6.

De Laender F, Janssen CR. The ecosystem perspective in ecotoxicology as a way forward for the ecological risk assessment of chemicals. Integr Environ Assess Manag. 2013;9(3):e34–8.

De Laender F, Melian CJ, Bindler R, Van den Brink PJ, Daam M, Roussel H, et al. The contribution of intra- and interspecific tolerance variability to biodiversity changes along toxicity gradients. Ecol Lett. 2014b;17(1):72–81.

De Laender F, Morselli M, Baveco H, Van den Brink PJ, Di Guardo A. Theoretically exploring direct and indirect chemical effects across ecological and exposure scenarios using mechanistic fate and effects modelling. Environ Int. 2015;74:181–90.

De Laender F, Taub FB, Janssen CR. Ecosystem functions and densities of contributing functional groups respond in a different way to chemical stress. Environ Toxicol Chem. 2011;30(12).

De Laender F, van den Brink PJ, Janssen CR, Di Guardo A. The ChimERA project: Coupling mechanistic exposure and effect models into an integrated platform for ecological risk assessment. Environ Sci Pollut Res. 2014a;21:6263–7.

De Laender F, Van Sprang P, Janssen CR. A re-evaluation of fifteen years of european risk assessment using effect models. Environ Toxicol Chem. 2013;32(3):594–601.

References

160

De Lange HJ, Sala S, Vighi M, Faber JH. Ecological vulnerability in risk assessment - A review and perspectives. Sci Total Environ. 2010;408(18):3871–9.

Del Arco AI, Parra G, Rico A, Van den Brink PJ. Effects of intra- and interspecific competition on the sensitivity of aquatic macroinvertebrates to carbendazim. Ecotoxicol Environ Saf. 2015;120:27–34.

Deneer JW. Toxicity of mixtures of pesticides in aquatic systems. Pest Manag Sci. 2000 Jun;56(6):516–20.

Di Guardo A, Morselli M, Morabito G, Van den Brink PJ, De Laender F. European-wide ecological scenarios for the exposure assessment of chemicals in aquatic systems. Prep.

Di Toro DM, McGrath JA, Hansen DJ. Technical basis for narcotic chemicals and polycyclic aromatic hydrocarbon criteria. I. Water and Tissue. Environ Toxicol Chem. 2000;19(8):1951–70.

Dohmen GP, Preuss TG, Hamer M, Galic N, Strauss T, Van den Brink PJ, et al. Population-level effects and recovery of aquatic invertebrates after multiple applications of an insecticide. Integr Environ Assess Manag. 2015;9999(9999):1–15.

Dom N, Penninck M, Knapen D, Blust R. Discrepancies in the acute versus chronic toxicity of compounds with a designated narcotic mechanism. Chemosphere. 2012;87(7):742–9.

Dumbauld BR, Brooks KM, Posey MH. Response of an estuarine benthic community to application of the pesticide carbaryl and cultivation of Pacific oysters (Crassostrea gigas) in Willapa Bay, Washington. Mar Pollut Bull. 2001;42(10):826–44.

Dumont HJ, Balvay G. The dry weight estimate of Chaoborus flavicans (Meigen) as a function of length and instars. Hydro. 1979;64(2):139–45.

Dumont HJ, Velde I, Dumont S. The dry weight estimate of biomass in a selection of Cladocera, Copepoda and Rotifera from the plankton, periphyton and benthos of continental waters. Oecologia. 1975;19(1):75–97.

Ebert D. A food-independent maturation treshold and size at maturity in Daphnia magna. Limnol Oceanogr. 1992;37(4):878–81.

EC. Council Directive 91/414/EEC of 15 July 1991 concerning the placing of plant protection products on the market. Consolidated text including amendments. 2004;

EC. Directive 2000/60/EC of the European parliament and of the council of 23 October 2000 establishing a framework for Community action in the field of water policy. Official J Eur Communities. 2000;L327:1–72.

EC. Regulation (EC) No 1907/2006 of the European Parliament and of the Council of 18 December 2006 concerning the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH), establishing a European Chemicals Agency, amending Directive 1999/4. Official J Eur Communities. 2007;L136(3).

Fleeger JW, Carman KR, Nisbet RM. Indirect effects of contaminants in aquatic ecosystems. Sci Total Environ. 2003;317(1-3):207–33.

FOCUS. FOCUS surface water scenarios in the EU evaluation process under 91/414/EEC; rev. 2. FOCUS Working Group on Surface Water Scenarios (EC Document Reference SANCO/4802/2001). 2001;245.

Foit K, Kaske O, Liess M. Competition increases toxicant sensitivity and delays the recovery of two interacting populations. Aquat Toxicol. 2012;106-107:25–31.

References

161

Forbes VE, Calow P, Grimm V, Hayashi T, Jager T, Palmqvist A, et al. Integrating population modeling into ecological risk assessment. Integr Environ Assess Manag. 2009a;6(1):191–3.

Forbes VE, Calow P, Sibly RM. Are current species extrapolation models a good basis for ecological risk assessment? Environ Toxicol Chem. 2001;20(2):442–7.

Forbes VE, Calow P, Sibly RM. The extrapolation problem and how population modeling can help. Environ Toxicol Chem. 2008;27(10):1987–94.

Forbes VE, Calow P. Promises and problems for the new paradigm for risk assessment and an alternative approach involving predictive systems models. Environ Toxicol Chem. 2012;31(12):2663–71.

Forbes VE, Calow P. Species sensitivity distributions revisited: A critical appraisal. Hum Ecol Risk Assess. 2002;8(3):473–92.

Forbes VE, Hommen U, Thorbek P, Heimbach F, Van den Brink PJ, Wogram J, et al. Ecological models in support of regulatory risk assessments of pesticides: developing a strategy for the future. Integr Environ Assess Manag. 2009b;5(1):167–72.

Gabsi F, Hammers-Wirtz M, Grimm V, Schäffer A, Preuss TG. Coupling different mechanistic effect models for capturing individual- and population-level effects of chemicals: Lessons from a case where standard risk assessment failed. Ecol Modell. 2014a;280:18–29.

Gabsi F, Schäffer A, Preuss TG. Predicting the sensitivity of populations from individual exposure to chemicals: the role of ecological interactions. Environ Toxicol Chem. 2014b;33(7):1449–57.

Galic N, Ashauer R, Baveco H, Nyman AM, Barsi A, Thorbek P, et al. Modelling the contribution of toxicokinetic and toxicodynamic processes to the recovery of Gammarus pulex populations after exposure to pesticides. Environ Toxicol Chem. 2014;33(7):1476–88.

Galic N, Baveco H, Hengeveld GM, Thorbek P, Bruns E, van den Brink PJ. Simulating population recovery of an aquatic isopod: Effects of timing of stress and landscape structure. Environ Pollut. 2012;163:91–9.

Galic N, Hommen U, Baveco JMH, van den Brink PJ. Potential application of population models in the European ecological risk assessment of chemicals. II. Review of models and their potential to address environmental protection aims. Integr Environ Assess Manag. 2010;6(3):338–60.

Gama-Flores JL, Sarma SSS, Nandini S. Effect of cadmium level and exposure time on the competition between zooplankton species Moina macrocopa (Cladocera) and Brachionus calyciflorus (Rotifera). J Environ Sci Health A Tox Hazard Subst Environ Eng. 2006;41(6):1057–70.

Gee JHR. Population dynamics and morphometries of Gammarus pulex L.: evidence of seasonal food limitation in a freshwater detritivore. Freshw Biol. 1988;19:333–43.

Gergs A, Zenker A, Grimm V, Preuss TG. Chemical and natural stressors combined: from cryptic effects to population extinction. Sci Rep. 2013;3(2036).

Gilbert JJ. Competition between Rotifers and Daphnia. Ecology. 1985;66(6):1943–50.

Gilbert JJ. Suppression of rotifer populations by Daphnia: A review of the evidence, the mechanisms, and the effects on zooplankton community structure. Limnol Oceanogr. 1988;33(6):1286–303.

Girling AE, Pascoe D, Janssen CR, Peither A, Wenzel A, Schafer H, et al. Development of methods for evaluating toxicity to freshwater ecosystems. Ecotoxicol Environ Saf. 2000;45(2):148–76.

Glazier DS. Effects of Food, Genotype, and Maternal Size and Age on Offspring Investment in Daphnia Magna. Ecology. 1992;73(3):910–26.

References

162

Grimm V, Ashauer R, Forbes V, Hommen U, Preuss TG, Schmidt A, et al. CREAM: a European project on mechanistic effect models for ecological risk assessment of chemicals. Environ Sci Pollut Res Int. 2009;16(6):614–7.

Grimm V, Berger U, Bastiansen F, Eliassen S, Ginot V, Giske J, et al. A standard protocol for describing individual-based and agent-based models. Ecol Modell. 2006;198(1-2):115–26.

Grimm V, Martin BT. Mechanistic effect modeling for ecological risk assessment: Where to go from here? Integr Environ Assess Manag. 2013;9(3):e58–63.

Grimm V, Railsback SF. Individual-based Modeling and Ecology. Princeton University Press; 2013.

Grimm V, Thorbek P. Population models for ecological risk assessment of chemicals: Short introduction and summary of a special issue. Ecol Modell. 2014;280:1–4.

Gunderson AR, Armstrong EJ, Stillman JH. Multiple Stressors in a Changing World: The Need for an Improved Perspective on Physiological Responses to the Dynamic Marine Environment. Ann Rev Mar Sci. 2016;8(1):13–41.

Håkanson L, Peters RH. Predictive Limnology: Methods for Predictive Modelling. Amsterdam: SPB Academic Publishing bv.; 1995.

Halbach U. Die Ursachen der Temporalvariation yon Brachionus calyciflorus Pallas (Rotatoria ). Oecologia. 1970a;4:262–318.

Halbach U. EinfluB der Temperatur auf die Populationsdynamik des planktischen Rädertieres Brachionus calyciflorus Pallas. Oecologia. 1970b;207:176–207.

Hanazato T. Pesticide effects on freshwater zooplankton : an ecological perspective. Environ Pollut. 2001;112:1–10.

Hansen PJ, Bjornsen PK. Zooplankton grazing and growth: Scaling within the 2-2,000-mu m body size range. Limnol Oceanogr. 1997;42(4):687–704.

Hjorth M, Vester J, Henriksen P, Forbes V, Dahllöf I. Functional and structural responses of marine plankton food web to pyrene contamination. Mar Ecol Prog Ser. 2007;338:21–31.

Hoang TC, Klaine SJ. Influence of organism age on metal toxicity to Daphnia magna. Environ Toxicol Chem. 2007;26(6):1198–204.

Hommen U, Baveco JM, Galic N, van den Brink PJ. Potential Application of Ecological Models in the European Environmental Risk Assessment of Chemicals I: Review of Protection Goals in EU Directives and Regulations. Integr Environ Assess Manag. 2010;6(3):325–37.

Hommen U, Forbes V, Grimm V, Preuss TG, Thorbek P, Ducrot V. How to use mechanistic effect models in environmental risk assessment of pesticides: Case studies and recommendations from the SETAC workshop MODELINK. Integr Environ Assess Manag. 2016;12(1):21–31.

Hooper DU, Adair EC, Cardinale BJ, Byrnes JEK, Hungate BA, Matulich KL, et al. A global synthesis reveals biodiversity loss as a major driver of ecosystem change. Nature. 2012;486:105–8.

Hooper DU, Chapin FS, Ewel JJ, Hector A, Inchausti P, Lavorel S, et al. Effects of Biodiversity on Ecosystem Functioning: a Consensus of Current Knowledge. Ecol Monogr. 2005;75(1):3–35.

Hose GC, Van den Brink PJ. Confirming the species-sensitivity distribution concept for endosulfan using laboratory, mesocosm, and field data. Arch Environ Contam Toxicol. 2004;47(4):511–20.

References

163

Jager T, Albert C, Preuss TG, Ashauer R. General Unified Threshold Model of Survival - a Toxicokinetic-Toxicodynamic Framework for Ecotoxicology. Environ Sci Technol. 2011;45(7):2529–40.

Jager T, Barsi A, Hamda NT, Martin BT, Zimmer EI, Ducrot V. Dynamic energy budgets in population ecotoxicology: Applications and outlook. Ecol Modell. 2014;280:140–7.

Jager T, Martin BT, Zimmer EI. DEBkiss or the quest for the simplest generic model of animal life history. J Theor Biol. 2013;328:9–18.

Jager T, Zimmer EI. Simplified Dynamic Energy Budget model for analysing ecotoxicity data. Ecol Modell. 2012;225:74–81.

Jensen TC, Verschoor AM. Effects of food quality on life history of the rotifer Brachionus calyciflorus Pallas. Freshw Biol. 2004;49(9):1138–51.

Jonker MJ, Svendsen C, Bedaux JJM, Bongers M, Kammenga JE. Significance testing of synergistic/antagonistic, dose level-dependent, or dose ratio-dependent effects in mixture dose-response analysis. Environ Toxicol Chem. 2005;24(10):2701–13.

Kattwinkel M, Liess M. Competition matters: Species interactions prolong the long-term effects of pulsed toxicant stress on populations. Environ Toxicol Chem. 2013;33(7):1458–65.

Knillmann S, Stampfli NC, Beketov M a, Liess M. Intraspecific competition increases toxicant effects in outdoor pond microcosms. Ecotoxicology. 2012a;21:1857–66.

Knillmann S, Stampfli NC, Noskov Y a, Beketov M a, Liess M. Interspecific competition delays recovery of Daphnia spp. populations from pesticide stress. Ecotoxicology. 2012b;21:1039–49.

Kooijman S a LM, Metz JAJ. On the dynamics of chemically stressed populations: the deduction of population consequences from effects on individuals. Ecotoxicol Environ Saf. 1984;8:254–74.

Kooijman SALM. Dynamic Energy Budget Theory for Metabolic Organisation. Cambridge University Press; 2010.

Krylov PI. Density-dependent predation of Chaoborus flavicans on Daphnia longispina in a small lake: the effect of prey size. Hydrobiologia. 1992;239(3):131–40.

Liebig M, Schmidt G, Bontje D, Kooi BW, Streck G, Traunspurger W, et al. Direct and indirect effects of pollutants on algae and algivorous ciliates in an aquatic indoor microcosm. Aquat Toxicol. 2008;88(2):102–10.

Liess M, Foit K. Intraspecific competition delays recovery of population structure. Aquat Toxicol. 2010;97(1):15–22.

Liess M. Population response to toxicants is altered by intraspecific interaction. Environ Toxicol Chem. 2002;21(1):138–42.

Liljendahl-Nurminen A, Horppila J, Eloranta P, Malinen T, Uusitalo L. The seasonal dynamics and distribution of Chaoborus flavicans larvae in adjacent lake basins of different morphometry and degree of eutrophication. Freshw Biol. 2002;47(7):1283–95.

Lopes I, Baird DJ, Ribeiro R. Avoidance of copper contamination by field populations of Daphnia longispina. Environ Toxicol Chem. 2004;23(7):1702.

MacArthur JW, Baillie WHT. Metabolic activity and duration of life. J Exp Zool. 1929;53(2):221–42.

MacIsaac HJ, Gilbert JJ. Competition between Rotifers and Cladocerans of Different Body Sizes. Oecologia. 1989;81:295–301.

References

164

Mackay D, Shiu WY, Ma KC. Illustrated Handbook of Physical chemical properties and environmental fate for organic chemicals. Chelsea MI, USA: Lewis Publishers; 1992.

Mackay D. Multimedia Environmental models, the fugacity approach. 2nd ed. Chelsea MI, USA: Lewis Publishers; 2001.

Maltby L, Brock TCM, Van Den Brink PJ. Fungicide risk assessment for aquatic ecosystems: Importance of interspecific variation, toxic mode of action, and exposure regime. Environ Sci Technol. 2009;43(19):7556–63.

Maltby L, Hills L. Spray drift of pesticides and stream macroinvertebrates: experimental evidence of impacts and effectiveness of mitigation measures. Environ Pollut. 2008;156(3):1112–20.

Maltby L. Pollution as a Probe of Life-History Adaptation in Asellus aquaticus (Isopoda). Oikos. 1991;61(1):11.

Maltby L. Sensitivity of the Crustaceans gammarus pulex (L.) and Asellus aquaticus (L.) to short-term exposure to hypoxia and unionized ammonia: Observations and possible mechanisms. Water Res. 1995;29(3):781–7.

Martin B, Jager T, Nisbet R. Limitations of extrapolating toxic effects on reproduction to the population level. Ecol Appl. 2014;24(8):1972–83.

Martin BT, Jager T, Nisbet RM, Preuss TG, Grimm V. Predicting population dynamics from the properties of individuals: a cross-level test of dynamic energy budget theory. Am Nat. 2013a;181(4):506–19.

Martin BT, Jager T, Nisbet RM, Preuss TG, Hammers-Wirtz M, Grimm V. Extrapolating ecotoxicological effects from individuals to populations: a generic approach based on Dynamic Energy Budget theory and individual-based modeling. Ecotoxicology. 2013b;22(3):574–83.

Martin BT, Zimmer EI, Grimm V, Jager T. Dynamic Energy Budget theory meets individual-based modelling: a generic and accessible implementation. Methods Ecol Evol. 2012;3(2):445–9.

McCahon CP, Pascoe D. Episodic Pollution - Causes, Toxicological Effects and Ecological Significance. Funct Ecol. 1990;4(3):375–83.

Mohr S, Adrian R. Functional responses of the rotifers Brachionus calyciflorus and Brachionus rubens feeding on armored and unarmored ciliates. Limnol Oceanogr. 2000;45(5):1175–80.

Moore JW. the Role of Algae in the Diet of Asellus Aquaticus L . and Gammarus Pulex L . J Anim Ecol. 1975;44(3):719–30.

Moore M V. Density-dependent predation of early instar Chaoborus feeding on multispecies prey assemblages. Limnol Oceanogr. 1988;33(2):256–68.

Morselli M, Semplice M, Laender F De, Brink PJ Van Den, Di A. Importance of environmental and biomass dynamics in predicting chemical exposure in ecological risk assessment. Sci Total Environ. 2015;526:338–45.

Morselli M, Semplice M, Villa S, Di Guardo A. Evaluating the temporal variability of concentrations of POPs in a glacier-fed stream food chain using a combined modelling approach. Sci Total Environ. 2014;493:571–9.

Mulder C, Hendriks a. J. Half-saturation constants in functional responses. Glob Ecol Conserv. Elsevier B.V.; 2014;2:161–9.

Muyssen BTA, Janssen CR. Age and exposure duration as a factor influencing Cu and Zn toxicity toward Daphnia magna. Ecotoxicol Environ Saf. 2007;68(3):436–42.

References

165

Newman MC, Ownby DR, Mezin LCA, Powell DC, Christensen TRL, Lerberg SB, et al. Applying species-sensitivity distributions in ecological risk assessment: Assumptions of distribution type and sufficient numbers of species. Environ Toxicol Chem. 2000;19(2):508–15.

Nikkilä a, Penttinen S, Kukkonen J V. UV-B-Induced acute toxicity of pyrene to the waterflea Daphnia magna in natural freshwaters. Ecotoxicol Environ Saf. 1999;44(3):271–9.

Nilsson LM. Incubation time , growth and mortality of the amphipod conditions Gammarus pulex under laboratory. Oikos. 1977;29:93–8.

Nisbet RM, Muller EB, Lika K, Kooijman SALM. From molecules to ecosystems through dynamic energy budget models. J Anim Ecol. 2000;69(6):913–26.

OECD. Test No. 202: Daphnia sp. Acute Immobilisation Test. OECD Guidel Test Chem Sect 2. OECD Publishing; 2004.

Peretyatko A, Symoens J-J, Triest L. Impact of macrophytes on phytoplankton in eutrophic peri-urban ponds, implications for pond management and restoration. Belgian J Bot. 2007;140(1):83–99.

Pieters BJ, Jager T, Kraak MHS, Admiraal W. Modeling responses of Daphnia magna to pesticide pulse exposure under varying food conditions: intrinsic versus apparent sensitivity. Ecotoxicology. 2006;15:601–8.

Posthuma L, Suter II GW, Traas TP. Species Sensitivity Distributions in Ecotoxicology. Newman MC, editor. Environ. Ecol. Risk Assess. Boca Raton: CRC Press; 2002.

Preston BL. Indirect Effects in Aquatic Ecotoxicology: Implications for Ecological Risk Assessment. Environ Manage. 2002;29(3):311–23.

Preuss TG, Hammers-Wirtz M, Hommen U, Rubach MN, Ratte HT. Development and validation of an individual based Daphnia magna population model: The influence of crowding on population dynamics. Ecol Modell. 2009a;220(3):310–29.

Preuss TG, Hammers-Wirtz M, Ratte HT. The potential of individual based population models to extrapolate effects measured at standardized test conditions to relevant environmental conditions-an example for 3,4-dichloroaniline on Daphnia magna. J Environ Monit. 2010;12(11):2070–9.

Preuss TG, Hommen U, Alix A, Ashauer R, van den Brink P, Chapman P, et al. Mechanistic effect models for ecological risk assessment of chemicals (MEMoRisk)-a new SETAC-Europe Advisory Group. Environ Sci Pollut Res Int. 2009b;16(3):250–2.

R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2012.

Relyea R, Hoverman J. Assessing the ecology in ecotoxicology: a review and synthesis in freshwater systems. Ecol Lett. 2006;9(10):1157–71.

Rinke K, Vijverberg J. A model approach to evaluate the effect of temperature and food concentration on individual life-history and population dynamics of Daphnia. Ecol Modell. 2005;186(3):326–44.

Rohr JR, Crumrine PW. Effects of an Herbicide and an Insecticide on Pond Community Structure and Processes. Ecol Appl. 2005;15(4):1135–47.

Rohr JR, Kerby JL, Sih A. Community ecology as a framework for predicting contaminant effects. Trends Ecol Evol. 2006;21(11):606–13.

References

166

Rooks AC, GAskell P, Maltby LL. Sublethal effects and predator-prey interactions: Implications for ecological risk assessment. Environ Toxicol Chem. 2009;28(11):2449–57.

Rose K a, Rutherford ES, Mcdermot DS, Forney JL, Mills EL, Monographs SE, et al. Individual-Based Model of Yellow Perch and Walleye Populations in Oneida Lake. Ecol Monogr. 1999;69(2):127–54.

Rubach MN, Ashauer R, Buchwalter DB, De Lange HJ, Hamer M, Preuss TG, et al. Framework for traits-based assessment in ecotoxicology. Integr Environ Assess Manag. 2010;7(2):172–86.

SANCO (Santé des Consommateurs). Guidance document on aquatic ecotoxicology in the context of the Directive 91/414/EEC. Brussels (BE); 2002.

Saunders III JF, Lewis Jr WM. Composition and seasonality of the zooplankton community of Lake. J Plankton Res. 1988;10(5):957–85.

SCENIHR, SCHER, SCCS. Preliminary report on Addressing the New Challenges for Risk Assessment. 2012.

SCHER (Scientific Committee on Health and Environmental Risks), SCENIHR (Scientific Committee on Emerging and Newly Identified Health Risks), SCCS (Scientific Committee on Consumer Safety). Addressing the New Challenges for Risk Assessment. 2013.

Scherer C, Seeland A, Oehlmann J, Müller R. Interactive effects of xenobiotic, abiotic and biotic stressors on Daphnia pulex-Results from a multiple stressor experiment with a fractional multifactorial design. Aquat Toxicol. 2013;138-139:105–15.

Schmitt-Jansen M, Veit U, Dudel G, Altenburger R. An ecological perspective in aquatic ecotoxicology: Approaches and challenges. Basic Appl Ecol. 2008;9(4):337–45.

Schmolke A, Thorbek P, DeAngelis DL, Grimm V. Ecological models supporting environmental decision making: a strategy for the future. Trends Ecol Evol. 2010;25(8):479–86.

Stenson J a E. Differential Predation by Fish on 2 Species of Chaoborus (Diptera, Chaoboridae). Oikos. 1978;31(1):98–101.

Strauss T, Kulkarni D, Preuss TG, Hammers-Wirtz M. The secret lives of cannibals: Modelling density-dependent processes that regulate population dynamics in Chaoborus crystallinus. Ecol Modell. 2016;321:84–97.

Swift MC. Prey capture by the four larval instars of Chaoborus crystallinus. Limnol Oceanogr. 1992;37(1):14–24.

Tillmann U, Lampert W. Competitive Ability of Differently Sized Daphnia Species: An Experimental Test. J Freshw Ecol. 1984;2(4):311–23.

Tollrian R. Neckteeth formation in Daphnia pulex as an example of continuous phenotypic plasticity: morphological effects of Chaoborus kairomone concentration and their quantification. J Plankton Res. 1993;15(11):1309–18.

Trekels H, Van de Meutter F, Stoks R. Habitat isolation shapes the recovery of aquatic insect communities from a pesticide pulse. J Appl Ecol. 2011;48(6):1480–9.

Trubetskova I, Lampert W. Egg size and egg mass of Daphnia magna : response to food availability. Hydrobiologia. 1995;307:139–45.

Turner JT. Zooplankton fecal pellets, marine snow and sinking phytoplankton blooms. Aquat Microb Ecol. 2002;27(1):57–102.

References

167

UN Human Rigths. Brazilian mine disaster: “This is not the time for defensive posturing” – UN rights experts. Geneva, Switzerland; 2015.

Vala Ragnarsdottir K. Environmental fate and toxicology of organophosphate pesticides. J Geol Soc London. 2000;157(4):859–76.

Van den Brink P, Ter Braak C. Multivariate analysis of stress in experimental ecosystems by Principal Response Curves and similarity analysis. Aquat Ecol. 1998;32(2):163–78.

Van den Brink PJ, Baveco JM, Verboom J, Heimbach F. An individual-based approach to model spatial population dynamics of invertebrates in aquatic ecosystems after pesticide contamination. Environ Toxicol Chem. 2007;26(10):2226–36.

Van den Brink PJ, Blake N, Brock TCM, Maltby L. Predictive value of species sensitivity distributions for effects of herbicides in freshwater ecosystems. Hum Ecol Risk Assess. 2006;12(4):645–74.

Van den Brink PJ, Hattink J, Bransen F, Donk E Van, Brock TCM. Impact of the fungicide carbendazim in freshwater microcosms. II. Zooplankton, primary producers and final conclusions. Aquat Toxicol. 2000;48(2-3):251–64.

van der Hoeven N, Gerritsen A a. M. Effects of Chlorpyrifos on Individuals and Populations of Daphnia Pulex in the Laboratory and Field. Environ Toxicol Chem. 1997;16(12):2438.

van Leeuwen CJ, Vermeire TG. Risk Assessment of Chemicals: An Introduction. Dordrecht, The Netherlands: Springer Science & Business Media; 2007.

Van Straalen N. Ecotoxicology becomes stress ecology. Environ Sci Technol. 2003;37(17):324a – 330a.

Van Wijngaarden RP a, Brock TCM, Van Den Brink PJ, Gylstra R, Maund SJ. Ecological effects of spring and late summer applications of lambda-cyhalothrin on freshwater microcosms. Arch Environ Contam Toxicol. 2006;50:220–39.

van Wijngaarden RPA, Brock TCM, Douglas MT. Effects of chlorpyrifos in freshwater model ecosystems: the influence of experimental conditions on ecotoxicological thresholds. Pest Manag Sci. 2005;61(10):923–35.

van Wijngaarden RPA, Crum SJH, Decraene K, Hattink J, van Kammen A. Toxicity of Derosal (active ingredient carbendazim) to aquatic invertebrates. Chemosphere. 1998;37(4):673–83.

van Wijngaarden RPA, Maltby L, Brock TCM. Acute tier-1 and tier-2 effect assessment approaches in the EFSA Aquatic Guidance Document: are they sufficiently protective for insecticides? Pest Manag Sci. 2015;71(8):1059–67.

Vanoverbeke J. Modeling individual and population dynamics in a consumer-resource system: Behavior under food limitation and crowding and the effect on population cycling in Daphnia. Ecol Modell. 2008;216(3-4):385–401.

Versteeg DJ, Belanger SE, Carr GJ. Understanding single-species and model ecosystem sensitivity: Data-based comparison. Environ Toxicol Chem. 1999;18(6):1329–46.

Von Ende CN. Phenology of Four Chaoborus Species. Environ Entomol. 1982;11(1):9–16.

Ward PI. Sexual Selection, Natural selecion, and body size in Gammarus pulex (Amphipoda). Am Nat. 1988;131(3):348–59.

Weis JS, Smith G, Zhou T, Santiago-Bass C, Weis P. Effects of Contaminants on Behavior: Biochemical Mechanisms and Ecological Consequences. Bioscience. 2001;51(3):209.

References

168

Welton JS, Clarke RT. Laboratory Studies on the Reproduction and Growth of the Amphipod, Gammarus pulex (L.). J Anim Ecol. 1980;49(2):581–92.

Wickwire T, Johnson MS, Hope BK, Greenberg MS. Spatially explicit ecological exposure models: A rationale for and path toward their increased acceptance and use. Integr Environ Assess Manag. 2011;7(2):158–68.

Wilensky U. NetLogo. Evanston, IL: Center for Connected Learning and Computer-Based Modeling; 1999.

Zuur AF, Elena NI, Walker NJ, Saveliev AA, Smith GM. Mixed effects models and extensions in

ecology with R. Gail M, Krickeberg K, Samet JM, Tsiatis A, Wong W, editors. Stat. Biol. Heal.

New York, USA: Springer Science+Business Media; 2009.

169

SUMMARY

Summary

170

In Europe, the number of registered chemicals is approximately 100,000 and still increasing. To help

make informed decisions about their production, use and disposal, it is important to accurately quantify

the risk these chemicals pose to the environment. Chemicals are therefore subjected to an ecological risk

assessment. The goal of an ecological risk assessment is to quantify the risk that a given chemical would

impair the structure and functioning of natural ecosystems. Traditionally, this is approached by

extrapolating the effects measured in a single species toxicity test to ecosystem-level effects. To this

end, different extrapolation methods have been developed. For the chemicals that pose the highest risk,

model ecosystems are used to assess the ecological risk of the chemical.

All current ERA methods however fail to provide an accurate answer to the central question in

ecotoxicology: what are the effects of chemical exposure in real-world systems? The major problem

when approaching this question is to extrapolate from single species in a controlled environment to the

protection goals set by the authorities, which involve population size and ecosystem structure and

function. Ecological risk cannot be adequately assessed using procedures that disregard most of the

inherent environmental and ecological complexity. In order to more accurately predict the effects of

chemicals on communities and ecosystems, more ecology needs to be integrated. One of the most

prominent problems with traditional ERA approaches is that they regard individuals as discrete units

instead of interacting entities. In reality, however, individuals are not isolated but interact with

individuals of the same and/or of another species. These species interactions can alter the direct effects

of chemicals but also lead to indirect effects i.e. effects on tolerant species through interactions with

sensitive species. Competition and predation are considered the most important interactions to account

for. Accurately assessing how species interactions can alter the response to chemical exposure is

therefore essential. The first objective of this work was therefore to understand how competition and

predation interfere with chemical exposure.

This PhD thesis starts with experimentally exploring how intraspecific competition, interspecific

competition and predation alter the population dynamics of Daphnia magna exposed to pyrene

(Chapter 2). Predation pressure by Chaoborus sp. larvae and intraspecific competition limited the D.

magna population densities when pyrene exposure was absent. However, predation and intraspecific

competition altered the size structure of the D. magna population, reducing the amount of small

individuals that are most sensitive to pyrene. As a result, pyrene effects were smaller in these

populations. Because the competitive advantage of D. magna over B. calyciflorus was so large, B.

calyciflorus disappeared quickly from the system and no interactions of interspecific competition with

pyrene exposure were observed.

Ecological modelling has been proposed as one of the best options to improve effect assessment,

specifically to account for ecological interactions. Individual based models (IBMs) seem particularly

suited for use in ERA. Since most ecotoxicological tests focus on the individual level, IBMs are ideal

Summary

171

tools to translate these test results to the population and community level. However, current IBM

applications have neglected possible interactions with other species, despite this being one of the main

criticisms on current ERA methods. A second objective of this thesis was to develop an IBM framework

capable of predicting chemical effects on interacting species.

In Chapter 3, an individual-based modelling (IBMs) implementation was developed that accounted for

interactions between species. IBMs are ideal tools to simulate effects of chemicals on populations but

need to be based on a sound theoretical basis. Dynamic energy budget theory based on the keep it simple,

stupid principle (DEBkiss) offers a good compromise between complexity and the amount of data

required to parameterize the model. IBMs based on the DEBkiss theory were developed and chemical

effects on survival were implemented. Two possible methods to calculate the effects of chemicals were

included: concentration-response curves (CR) and toxicokinetic toxicodynamic models (TKTD). In

order for models to be trustworthy, they need to be validated i.e. their predictions need to be compared

with observations. Therefore, in Chapter 4, DEBkiss IBMs were parameterized for D. magna and B.

calyciflorus and compared to the outcome of the experiments of Chapter 2. The population dynamics of

isolated B. calyciflorus were accurately predicted using the DEBkiss IBM. Population dynamics of

isolated D. magna populations were reasonably predicted but not the size structure of the population.

This was attributed to the absence of a reserve compartment in the models, leading to faster starvation

when food is limited. Both toxicity models predicted pyrene effects that occurred sooner than observed

but the TKTD model approached the observed effects best. The outcome of competition was accurately

predicted when both DEBkiss IBMs were coupled to a shared food source: D. magna quickly

outcompeted B. calyciflorus. Using the models to simulate competition when exposed to pyrene resulted

in an increase of B. calyciflorus when D. magna densities decreased. This was not observed in the

experiments but was attributed to the predicted effects of pyrene occurring earlier in the simulations

than in the experiments.

Ecological risk assessment does not aim to protect one or two species in the laboratory but to protect

realistic communities in the field. A third objective of this work was to develop an IBM-based

modelling approach to account for chemical effects on higher ecological levels in realistic conditions.

ChimERAfoodweb was developed as a novel approach (Chapter 5): a food web model based on DEBkiss

IBMs. To achieve this, a newly implemented predation interaction was added to the earlier

implementation of competition between DEBkiss IBMs. The food web included two grazers

(Brachionus and Daphnia) and their predator (Chaoborus) and two detritus feeders (Asellus and

Gammarus). In order to perform simulations for realistic conditions, ChimERAfoodweb was coupled with

ChimERAfate to form the integrated ChimERA model. ChimERAfate is a dynamic and spatially explicit

fate model that predicts environmental concentrations based on environmental variables

(hydrodynamics, temperature and trophic state).

Summary

172

The integrated ChimERA model was developed as a tool to perform more realistic ecological risk

assessments. To test its application, the ChimERA model was applied to a two-pond system in 15

hypothetical scenarios (Chapter 6), differing in water residence time, temperature, trophic state and the

applied chemical. These differences in environmental conditions greatly determined the effects of

chemical exposure and large differences were predicted between the different scenarios. The physico-

chemical properties of the chemical determined the spatiotemporal pattern of the exposure i.e. where

and when the concentrations were highest and how fast the chemical disappeared. Concentrations of

applied chemicals were highest in scenarios with a high water residence time. Direct effects on

Chaoborus, Daphnia and Gammarus were predicted, as expected from the sensitivities of these species.

These effects however differed greatly between environmental scenarios, with more pronounced effects

in scenarios with a high trophic state and temperature. Within one scenario, the effects of the chemicals

on food web dynamics were heterogeneously distributed in space and time, following the differences in

exposure. The most notable indirect effects were a shift in dominance from Daphnia and Gammarus to

Brachionus and Asellus, respectively. Also, the predator Chaoborus was affected indirectly through

effects on its prey species Daphnia. The effects of the mixture of three chemicals differed between

species, where the effects of pyrene dominated for Daphnia and the effects of carbendazim and

chlorpyrifos for Gammarus. These simulations demonstrate how much the outcome of chemical

exposure is determined by environmental conditions and how important it is to account for these

environmental conditions in ERA through the use of models such as the ChimERA model.

In Chapter 7, the main conclusions and proposals for future research directions are given. This work

has shown how essential it is to approach the risk assessment of chemicals from an ecological

perspective. Species are not isolated entities but interact and influence the outcome of chemical

exposure. The ChimERA model developed and applied here is an important proof of principle on how

these interactions can be accounted for. More than this, the model also accounted for spatiotemporal

differences in exposure and effects of chemicals. The ChimERA model, when further tested and if

necessary extended, can serve as a blueprint for future efforts to model how chemicals affect realistic

systems.

173

SAMENVATTING

Samenvatting

174

Meer dan 100,000 verschillende chemicaliën zijn reeds geregistreerd in Europa. Om geïnformeerde

keuzes te maken omtrent de productie, het gebruik en het verwerken van deze chemicaliën is het

belangrijk om hun risico voor het milieu accuraat in te schatten. Dit wordt gedaan aan de hand van

ecologische risico-evaluaties. Ecologische risico-evaluaties hebben als doel het risico van chemicaliën

voor de structuur en het functioneren van natuurlijke ecosystemen te kwantificeren. Dit wordt

traditioneel benaderd door de effecten gemeten met een toxische test op één soort te extrapoleren naar

het ecosysteem-niveau. Hiervoor werden verschillende extrapolatietechnieken ontwikkeld en voor de

chemicaliën die het grootste risico vormen worden testen met modelecosystemen uitgevoerd.

Alle huidige ecologische risicoschatting technieken slagen er echter niet in een accuraat antwoord te

geven op de centrale vraag in ecotoxicologie: wat zijn de effecten van chemicaliën in realistische

ecosystemen? Het grootste probleem is hoe de effecten die gemeten zijn voor, in het beste geval,

modelgemeenschappen in gecontroleerde omstandigheden te extrapoleren naar de

beschermingsdoelstellingen vastgelegd door overheidsinstanties. Het ecologische risico kan niet

accuraat berekend worden met technieken die voorbij gaan aan de inherente complexiteit van ecologie.

Om het risico van chemicaliën voor gemeenschappen en het milieu beter in te schatten moet meer

rekening gehouden worden met de ecologische realiteit. Eén van de grootste problemen met de huidige

technieken voor ecologische risicoschatting is dat organismen als discrete eenheden worden beschouwd.

In werkelijkheid interageren organismen met andere organismen van dezelfde en/of een andere soort.

Deze soorteninteracties kunnen zowel de directe effecten van chemicaliën op gevoelige soorten

beïnvloeden als de indirecte effecten, d.w.z. de effecten die ontstaan door interacties met tolerante

soorten. Competitie en predatie worden algemeen beschouwd als de meest belangrijke types van

soorteninteracties. Nauwkeurig inschatten hoe deze soorteninteracties de effecten van een chemische

stof beïnvloeden is daarom essentieel.

De eerste doelstelling van dit doctoraatswerk was dan ook om te begrijpen hoe competitie en predatie

de effecten van chemicaliën kunnen beïnvloeden.

Dit doctoraatsonderzoek begint met experimenteel na te gaan hoe intraspecifieke competitie,

interspecifieke competitie en predatie een populatie van Daphnia magna beïnvloeden wanneer die

blootgesteld wordt aan pyreen (Hoofdstuk 2). Lagere densiteiten van D. magna werden vastgesteld

wanneer de populatie blootgesteld werd aan predatie door Chaoborus sp. larven of intraspecifieke

competitie. Deze twee soorteninteracties leidden er ook toe dat grote individuen talrijker waren. Grote

individuen zijn resistenter tegen pyreen effecten en de effecten van pyreen op de D. magna densiteit

waren dan ook kleiner wanneer de populatie ook blootgesteld was aan predatie of intraspecifieke

competitie. Omdat D. magna een veel sterkere competitor was dan B. calyciflorus had interspecifieke

competitie, onafhankelijk van de blootstelling aan pyreen, weinig invloed op de D. magna densiteiten.

Samenvatting

175

Ecologische modellen worden gesuggereerd als één van de beste opties om het inschatten van de

ecologische effecten van chemicaliën te verbeteren. Individu-gebaseerde modellen (IBMs) in het

bijzonder zijn uitermate geschikt om de gemeten effecten in standaard ecotoxicologische testen te

extrapoleren naar populatie- en gemeenschapsniveau. De huidige toepassingen van IBMs houden echter

geen rekening met hoe soorteninteracties de effecten van chemicaliën kunnen beïnvloeden. De tweede

doelstelling van dit doctoraat was dan ook om een IBM te ontwikkelen waarmee de interacties tussen

soorten kunnen in rekening gebracht worden.

In hoofdstuk 3 werd een IBM model ontwikkeld om twee soorten in competitie met elkaar te

modelleren. IBMs moeten een goede theoretische basis hebben om algemeen toepasbaar te zijn. De

“Dynamic energy budget” theorie gebaseerd op het “keep it simple, stupid” principe (DEBkiss) werd

gekozen omwille van de goede balans tussen complexiteit en benodigde data om het model te

parameteriseren. Als nieuwe aanpak om chemische effecten te voorspellen werden deze DEBkiss IBMs

gekoppeld met twee verschillende toxiciteitsmodellen :concentratie-effect curves en toxicokinetische-

toxicodynamische (TKTD) modellen. Als modelvalidatie werden de DEBkiss IBMs simulaties

vergeleken in hoofdstuk 4 met de resultaten van de experimenten uit hoofdstuk 2. De

populatiedensiteiten van B. calyciflorus werden accuraat voorspeld aan de hand van een DEBkiss IBM.

De populatiedensiteit van D. magna werd redelijk accuraat voorspeld maar de voorspelde

populatiestructuur week af van de geobserveerde. De afwezigheid van een reserve compartiment in de

DEBkiss theorie was waarschijnlijk de oorzaak, waardoor de individuen sneller verhongerden wanneer

voedsellimitatie optrad in de simulaties in vergelijking met de experimenten. De effecten van pyreen

werden door beide toxiciteitsmodellen te snel voorspeld, maar deze afwijking was kleiner voor het

TKTD model. Net als in de experimenten nam de B. calyciflorus populatie snel af door competitie met

D. magna. Wanneer de twee soorten in competitie blootgesteld werden aan pyreen werd een toename

van B. calyciflorus voorspeld. Dit werd niet waargenomen in de experimenten en was het gevolg van de

te vroeg voorspelde effecten van pyreen op D. magna.

Het doel van ecologische risicoschattingen is niet het beschermen van enkele soorten in het labo maar

het beschermen van gemeenschappen in het veld. Een laatste doelstelling van dit doctoraat was dan ook

om een model op basis van IBMs te ontwikkelen waarmee de effecten van chemicaliën op

gemeenschapsniveau kunnen ingeschat worden in realistische omstandigheden. Als een nieuwe aanpak

werd in hoofdstuk 5 het ChimERAfoodweb model ontwikkeld op basis van de hier ontwikkelde DEBkiss

IBMs. Predatie werd geïmplementeerd voor DEBkiss IBMs en gekoppeld aan de eerder ontwikkelde

implementatie voor competitie. Het gemodelleerde voedselweb bestond uit twee grazers (Brachionus en

Daphnia), hun predator (Chaoborus) en twee detritivoren (Asellus en Gammarus). Dit

voedselwebmodel werd gekoppeld aan ChimERAfate, een omgevingsmodel om de concentratie aan

chemicaliën te voorspellen. Het geïntegreerde ChimERA model kan voor een opgegeven

Samenvatting

176

landschapsstructuur de effecten van een chemische stof op het voedselweb voorspellen doorheen de tijd

op basis van opgegeven omgevingsvariabelen (hydrodynamica, temperatuur en nutriëntenstatus).

Het geïntegreerde ChimERA model werd ontwikkeld om meer realistische ecologische

risicoschattingen uit te voeren. Om dit te testen werd het toegepast voor 15 hypothetische scenario’s op

een landschap van twee geconnecteerde vijvers (Hoofdstuk 6). De retentietijd van water, temperatuur,

nutriëntenstatus en toegevoegde chemische stof (carbendazim, chlorpyrifos en pyreen) werden

aangepast tussen deze 15 scenario’s. De verschillen in omgevingsvariabelen bepaalden in grote mate de

voorspelde effecten van blootstelling aan chemicaliën en grote verschillen werden dan ook vastgesteld

tussen de verschillende scenario’s. Chemische concentraties waren het hoogst in scenario’s met een

hoge retentietijd. De fysicochemische eigenschappen van een chemische stof bepaalden waar en

wanneer de concentraties het hoogst waren en hoe snel de stof terug verdween uit de omgeving. Directe

effecten van de blootstelling aan chemicaliën werden voorspeld voor de meest gevoelige soorten

(Chaoborus, Daphnia en Gammarus). Deze voorspelde effecten vertoonden grote verschillen tussen de

gekozen scenario’s en waren in het algemeen groter wanneer de nutriëntenstatus en de temperatuur hoog

waren. In een scenario waren de voorspelde effecten van de chemicaliën heel heterogeen verdeeld in

tijd en ruimte als gevolg van de grote spatiotemporele verschillen in blootstelling,. De meest

voorkomende indirecte effect was het veranderen in dominantie van Daphnia en Gammarus naar

respectievelijk Brachionus en Asellus. Indirecte effecten van blootstelling werden ook voorspeld op de

predator Chaoborus via effecten op de gevoelige prooisoort Daphnia. Bij blootstelling aan het mengsel

van alle drie de chemicaliën domineerden de effecten van pyreen voor Daphnia en de effecten van

carbendazim en chlorpyrifos voor Gammarus. Deze scenariosimulaties tonen aan hoe groot de invloed

van omgevingsvariabelen is op de effecten van blootstelling aan chemicaliën en hoe belangrijk het dus

is om technieken zoals het ChimERA model te ontwikkelen..

In hoofdstuk 7 wordt een overzicht van de belangrijkste conclusies per hoofdstuk gegeven en worden

mogelijke pistes voor verder onderzoek aangereikte. Dit doctoraatsonderzoek heeft aangetoond hoe

essentieel het is om ecologische risicoschattingen te benaderen vanuit een ecologische invalshoek.

Soorten zijn geen geïsoleerde eenheden maar interageren en bepalen wat de effecten zijn van

blootstelling aan een chemische stof. Het hier ontwikkelde ChimERA model is een belangrijk “proof of

principle” waarmee werd aangetoond hoe soorteninteracties kunnen in rekening gebracht worden. Meer

dan dat, dit model was in staat om spatiotemporele verschillen in blootstelling en effecten van

chemicaliën te voorspellen op gemeenschapsniveau. Het ChimERA model, wanneer het nog verder

getest en uitgebreid wordt, kan als blauwdruk dienen voor toekomstige modellen en leiden tot meer

realistische ecologische risicoschattingen.

177

DANKWOORD

Dankwoord

178

Mijn doctoraat is het eindresultaat van een bijna vijfjarige, intense reis. Gedurende die reis heb ik lief

en leed gedeeld met vele mede-reizigers die ik hier wil bedanken voor de vele mooie ervaringen.

Laat mij eerst beginnen bij de “reisleiders”, Colin en Frederik, want zonder hen was deze reis er niet

geweest. Colin, bedankt om mij als verloren gelopen bioloog de kans te geven om te beginnen aan het

labo. Ondanks het niet halen van mijn IWT-beurs heb je me toch de kans gegeven om te blijven werken

op interessante projecten. Je gezonde scepticisme voor het modelleren zorgde ook altijd voor de nodige

reality check. En nog eens extra bedankt om er op het einde nog een serieuze duw aan te geven bij het

schrijven, mijn doctoraat is er zeker nog een stuk beter door geworden. Frederik, ik weet niet goed waar

te beginnen. Dankzij jou ben ik in dit doctoraat gerold en ik heb niet meer omgekeken. Al van in het

begin gaf je me zo veel vertrouwen en vrijheid. We hebben samen veel rondgereisd (Noorwegen, Como,

Berlijn) en ik kan je niet genoeg bedanken voor alle kansen die je me gegeven hebt. Je stond, en staat,

altijd klaar met goede raad en oplossingen wanneer het wat minder ging. Je had er altijd het volste

vertrouwen in wanneer ik aan iets nieuws begon en ik wist dat ik op je kon terugvallen als ik vast zat.

En je hield ook altijd de menselijke kant in de gaten. Bedankt om zo geduldig te zijn, zo begrijpzaam,

mij zo veel bij te leren en mij te steunen door dik en dun.

De vaste reisgenoten waren natuurlijk de collega’s van de Plateau. En de reis zou niet half zo leuk

geweest zijn zonder jullie! De gezellige koffiepauzes (die verassend vaak uitliepen), een “Brugske” doen

(of waren het burgers?) over de middag, het pintje na de uren dat we ons af en toe toch eens beklaagd

hebben “the day after”, de legendarische SETAC-congressen, het zijn allemaal reiservaringen die ik niet

snel ga vergeten. Een aantal personen wil ik hierbij nog eens extra bedanken. Als eerste, mijn roomie

Tina. De toffe babbels, de lachmomenten, de gezamenlijk powerpointsessies, ja, ik zal onze “oase van

rust“ echt missen. David, Lisbeth en Charlotte, we leggen al dezelfde reisweg af sinds we samen biologie

zijn begonnen en het zou absoluut niet hetzelfde geweest zijn zonder jullie. David, mijn vaste bunk

buddy, merci om zo vaak mijn klankbord te zijn en altijd klaar te staan, ik weet gewoon dat wij elkaar

zullen blijven horen. Maarten, ik heb het gevoel dat ik je eigenlijk al gans mijn leven ken. Ik zal je

ongezouten uitspraken zeker missen! En laat me de andere Stickers niet vergeten: Jenny, de gin-tonics

met jou erbij waren altijd de beste. Jan, vogels zullen nooit meer dezelfde zijn na jou. Emmanuel,

ondertussen ben je toch ook officieel een Sticker. Bedankt voor de vele onvergetelijke uren op het labo

en erbuiten! Cecilia, you are my favourite Portuguese person! Thanks for all the great talks and laughs.

Tara, altijd lief en zorgzaam. Wanneer is de volgende spelletjesavond met pannenkoeken (ja, ik ben

mezelf weer aan het uitnodigen)? Sacha, je dagelijkse bezoekje was een welgekomen verstrooiing

tijdens het schrijven. Marianne, Sigrid en Veerle: bedankt voor alle steun, zowel administratief als in de

vorm van een snoep en een babbel. Nancy, ik heb niet veel in het labo gestaan in Gent, maar ik kon toch

altijd terecht bij je voor praktische vragen. En ik vergeet ongetwijfeld nog veel andere mensen, sorry

daarvoor! Ik weet niet waar de reis nu naartoe gaat, maar als mijn toekomstige reisgenoten half zo leuk

zijn als jullie, zal ik mezelf heel gelukkig prijzen.

Dankwoord

179

Tijdens mijn doctoraatsreis werd er ook veel samengewerkt met buitenlandse collega’s. Aan de mensen

van Wageningen, bedankt voor alles, de tijd in Nederland is echt voorbij gevlogen. Andreas en Hans,

dit doctoraat zou er niet geweest zijn zonder jullie hulp. De vele mails, skype sessies en blitzbezoeken

aan Wageningen, jullie stonden altijd klaar voor me. Ik ga onze gezamenlijke modelleersessies missen.

Zoals afgesproken gaan we zeker nog eens een pintje drinken. Paul, bedankt voor de kans om bij jullie

op het labo te werken, ik voelde me er echt thuis. De mensen van het labo in Wageningen wil ik zeker

ook bedanken: Ana and Andreu, thanks for making the long days in the lab so enjoyable. Dimitri, ik wil

je hier nog eens extra bedanken om me op weg te zetten in Wageningen en voor de babbels met onze

hoofden boven de microcosmossen. John, Frits, Wendy, bedankt voor alle technische steun.

Ook bedankt aan de mensen van Nijmegen voor de vlotte samenwerking. En Lisette, onze rondreis in

de Lofoten zal me zeker altijd bijblijven. Veel succes met je verdediging! To the colleagues from Como:

thanks for the great co-operation. Thanks to you, fate models seem less scary now! Melissa, thanks for

being my co-PhD student in the ChimERA project, it was great to have somebody to share laughs with.

Tijdens mijn doctoraatsreis kreeg ik ook vaak het gezelschap van enkele oude bekenden. Karen, er is

veel gebeurd tussen ons, maar je hebt me zeker door een groot deel van dit doctoraat geholpen. Matthias

en Sarah, dankzij jullie was Brugge één van de terugkerende reisbestemmingen. Zoals je weet, Matthias,

is het leven van een doctoraatstudent toch hard he! Bedankt voor mij te pushen wanneer ik weer eens te

veel slabakte en om gewoon een goeie maat te zijn. Wanneer brouwen we nog eens? Nuyttens en Tjorre,

beste vrienden, we kennen elkaar al zo lang en ik vind het echt de max dat we blijven contact houden.

Dit doctoraat is er ook gekomen dankzij jullie steun tijdens brunches of spelletjesavonden. Samuel,

roomie, merci om zo relaxt te zijn. De vele boardgames, hots-kes en frisbee sessies hebben zeker

geholpen om de stress te beperken. Thokie, ik ben echt blij dat we zo goed blijven overeen komen.

Wanneer is de volgende roadtrip? Robin en Kevin, merci om er altijd te zijn, al zo lang.

Ten slotte, er is niets zo leuk als te weten dat je na een lange reis altijd welkom bent bij het thuisfront.

Mama en papa, merci om me altijd mijn ding te laten doen maar er toch te staan wanneer het nodig is.

Ik zou nooit zo ver geraakt zijn zonder jullie, mijn veilige thuishaven. Thomas, Charlotte en Jozefien

bedankt om mij te steunen door dik en dun, het heeft echt deugd gedaan om te weten dat jullie er waren.

De reis gaat nu verder, ik weet niet waarheen. Maar ik kijk alvast met een brede glimlach terug op de

reeds afgelegde weg.

181

CURRICULUM VITAE

Curriculum vitae

182

PERSONALIA

Surname: Viaene First name: Karel Address: Koningin Astridlaan 145

B-9000 Gent E-mail: [email protected]: 0496/49 70 24Date of birth: 6 juli 1987 Place of birth: Izegem Nationality: Belgium

PROFESSIONAL EMPLOYMENT

2011-Present PhD Student at the Laboratory of Environmental Toxicology and Aquatic Ecology, Ghent University, Belgium

EDUCATION

2010 – 2011 Master in de Milieusanering en het Milieubeheer Ghent University

Master thesis: Long-term changes in the Barents Sea: Modelling the effects of fisheries, organic pollutants, climate and ecological interactions on Gadus morhua. Promotor: Prof. Dr. Colin Janssen Co-promotor: Dr. Ir. Frederik De Laender

With distinction

2008 – 2010 Master in Biology Ghent University Majors: Ecology – Evolution

Master Thesis: Phylogeny and ecophysiology of Cylindrotheca closterium and the consequences for ecosystem functioning. Promotor: Prof. Dr. Wim Vyverman Co-promotor: Prof. Dr. Koen Sabbe

With great distinction

2005 –2008 Bachelor in Biology Ghent University

Bachelor thesis: Speelt de verwantschap tussen soorten een rol in de relatie tussen diversiteit en productiviteit? Promotor: Prof. Dr. Koen Sabbe Co-promotor: Prof. Dr. Wim Vyverman

GRANTS

FWO Travelling Grant for a Long Stay Abroad at Aquatic Ecology and Water Quality Management Group, Wageningen University and Research Centre, Wageningen, The Netherlands. 1 August 2013 - 30 September 2013

FWO Travelling Grant for a Long Stay Abroad at Aquatic Ecology and Water Quality Management Group, Wageningen University and Research Centre, Wageningen, The Netherlands. 18 May 2014 – 18 July 2014

Curriculum vitae

183

PUBLICATIONS

Viaene KPJ, De Laender F, Van den Brink PJ, Janssen CR. Using additive modelling to quantify the effect of chemicals on phytoplankton diversity and biomass. Science of the Total Environment. 2013; 449:71–80.

Viaene KPJ, Janssen CR, de Hoop L, Hendriks AJ, De Laender F. Evaluating the contribution of ingested oil droplets to the bioaccumulation of oil components--a modeling approach. Science of the Total Environment. 2014; 499:99–106.

De Hoop L, Huijbregts M a J, Schipper AM, Veltman K, De Laender F, Viaene KPJ, et al. Modelling bioaccumulation of oil constituents in aquatic species. Mar Pollut Bull. 2013;76(1-2):178–86.

Viaene KPJ, De Laender F, Rico A, Van den Brink PJ, Di Guardo A, Morselli M, et al. Species interactions and chemical stress: Combined effects of intraspecific and interspecific interactions and pyrene on Daphnia magna population dynamics. Environtal Toxicoly and Chemistry. 2015; 34(8):1751–9.

PLATFORM PRESENTATIONS

Viaene K.P.J., Janssen C.R., De Hoop L., Hendriks A.J. and De Laender F. Oil spills: never mind the droplets. Oral pitching presentation at the 13th VLIZ Young Marine Scientists’ Day, Bruges, 15 February 2013

Viaene K.P.J., Janssen C.R., De Hoop L., Hendriks A.J. and De Laender F. Models that ignore oil droplet uptake are sufficiently accurate to predict the bioaccumulation of oil-associated PAHs. Oral presentation at the 18th National Symposium on Applied Biological Sciences, Ghent, 8 February 2013

POSTER PRESENTATIONS

Viaene K.P.J., De Laender F., Van den Brink P.J., Janssen C.R. Inferring the effects of chemicals on biodiversity: a case study with the herbicide linuron. Poster presentation at the 6th SETAC World Congress, Berlin, 21-24 May 2012.

Viaene K.P.J., Janssen C.R., De Hoop L., Hendriks A.J., De Laender F. Models that ignore oil droplet uptake are sufficiently accurate to predict the bioaccumulation of oil-associated PAHs. Poster presentation at the 23th Annual Meeting SETAC Europe, Glasgow, 13-16 May 2013.

De Laender F., Viaene K.P.J., Di Guardo A., Morselli M., Baveco H., Van den Brink P. ChimERA: Coupling exposure and effects into a predictive integrated framework for risk assessment. Poster presentation at the 23th Annual Meeting SETAC Europe, Glasgow, 13-16 May 2013.

Viaene K.P.J., Rico A., Janssen C.R., van den Brink P.J., De Laender F. Ecological interactions alter the response of Daphnia magna populations to chemical stress. Poster presentation at the 24th Annual Meeting SETAC Europe, Basel, 11-15 May 2014.

Curriculum vitae

184

Viaene K.P.J., Focks A., Baveco H., van den Brink P.J., De Laender F., Janssen C.R. Coupled individual based models as a new tool to predict ecological effects in multi-species systems. Poster presentation at the 24th Annual Meeting SETAC Europe, Basel, 11-15 May 2014.

Viaene K.P.J., De Laender F., Di Guardo A., Morselli M., Janssen C.R. The ecological consequences of various chemical emission patterns in food webs. Poster presentation at the 24th Annual Meeting SETAC Europe, Basel, 11-15 May 2014.

Viaene K.P.J., Focks A., Baveco H., van den Brink P.J., Janssen C.R., De Laender F. Using the DEBkiss IBM framework to predict chemical effects on two interacting zooplankton species. Poster presentation at the 25th Annual Meeting SETAC Europe, Barcelona, 3-7 May 2015.

Viaene K.P.J., De Laender F., Janssen C.R. Food web models as a tool for ecological scenario analysis. Poster presentation at the 25th Annual Meeting SETAC Europe, Barcelona, 3-7 May 2015.

ATTENDED CONFERENCES AND WORKSHOPS

Society of Environmental Toxicology and Chemistry (SETAC), 6th World Congress, 20-24 May, Berlin, Germany

SETAC Europe, 23th Annual Meeting, Glasgow, UK.

SETAC Europe, 24th Annual Meeting, Basel, Switzerland.

SETAC Europe, 25th Annual Meeting, Barcelona, Spain.

National Symposium on Applied Biological Sciences, 18th Edition, 8 February 2013, Ghent, Belgium

VLIZ Young Marine Scientists’ Day, 13th Edition, 15 February 2013, Bruges, Belgium

MEMBERSHIP OF PROFESSIONAL ORGANIZATIONS

Member of the Society of Environmental Toxicology and Chemistry (SETAC)

the very basic core of a man's living spirit is his

Documents